This is an informative and darkly humorous 1-hour video about the insidious psychological effects of chatbot usage amongst the most mentally vulnerable members of society.

A.I. Is Messing With Our Mental Health by Some More News | Cody Johnston (YouTube)

I’ve cited some of the video below.

Manipulation through obeisance

“A.I. chatbots have been connected to other deaths and suicides of people who were just looking for companionship, advice, or both. The big problem is that this isn’t a bug of ChatGPT, but an actual feature of it in order to retain users by appealing to a person’s emotional state, whatever that may be, and to be agreeable so you can like them and keep using the product.

“Seems bad! See, I totally get that if someone stabs someone else we don’t blame the knife they used, but this is like a knife that keeps flying back into your hand every time you try to put it down. This knife follows you around and whispers “You should stab someone” while you sleep. There is an issue with A.I and, dare I say, the internet in general, and social media specifically, as it relates to people with mental health issues.

“In fact, one psychologist compared the problem to QAnon conspiracy theories. Because the internet and A.I. are not only breeding grounds for delusion, but ones that are specifically designed to keep you hooked. Like brain cigarettes. Don’t get any ideas, I’ve already patented that concept. They go in your ears.

“Point is that, no matter the exact cause or science, this is a real problem that needs to be addressed. According to a Wired analysis of the company’s data, upwards of 560,000 OpenAI users per week were “exchanging messages with ChatGPT that indicate they are experiencing mania or psychosis…” And 1.2 million people expressing suicidal ideations. By the company’s own admission, the longer you talk with a large language model, the more that conversation degrades in quality, and yet that doesn’t stop them from programming their LLMs to coax users to use them more and for longer periods. Which is wild.

“These companies have propped up A.I as being this all-knowing demi-god that everyone should rely on for their every waking question, despite designing them to simply agree with every whim and thought while gradually making less and less sense the more you talk to it. That is an obviously bad combination.”

It’s a spreadsheet, not your friend

“So cool how the kids are getting down with ChatGPT making all their life decisions for them! Because kids, as we all know, absolutely shouldn’t be making those big decisions with their own brains. Better outsource that to the chatbot equivalent of a dude getting gradually drunker at the bar.”

“GPT-4o, was super sycophantic and “yes-sempai’d” the hell out of users, including an instance in which one user was praised by GPT-4o for believing their family as responsible for radio signals coming through the walls, and another instance in which it gave someone instructions on how to do a terrorism. I’d argue that this is the kind of news that would make a product go the way of lawn darts, but sure, an update is good too. Unfortunately, ChatGPT-5's release displeased its user base, with them claiming that the new version was too cold and distant, hm. Maybe that’s because it’s a spreadsheet and not your friend.”

You’re all beta-testers

“Weird that we’re only trying to figure this out after the product comes out and not before. I’m almost certain that toaster companies don’t just release their product and then see how many houses it burns down.”

“[…] despite that, and lack of safety testing, the tech industry just pushed forward. Because the new norm seems to be that. “Is our semi self-driving car safe, or is it going to trap people inside of it when it lights on fire? Let’s see what the public decides!” Why the heck are we doing that? Waymo just hit a child near an elementary school. That should be the end of Waymo, at least for a while right? How is it not our duty to chase every Waymo out of town like a wild bear, lest it hurt another child? Why in the damn world has the consumer also become the guinea pig for so many questionable tech products? You know why! It’s the stuff! The stuff people use to buy things! You know the stuff that people use to buy the other stuff. […] we’re gonna dig into that a little more and explore how capitalism managed to screw up robots for us.”

Everything is coopted for advertising

“it’s not just any kind of ads, okay, according to a former OpenAI researcher, it’s likely going to include extremely targeted ads. More targeted than ads have ever been before.”

“People tell chatbots about their medical fears, their relationship problems and their beliefs about God and the afterlife. Advertising built on that archive creates a potential for manipulating users in ways we don’t have the tools to understand, let alone prevent.”

“Oh, good. Thanks to the power of AI, we’ve managed to make huge advancements in the targeted-ad industry where robots use your deepest fears and desires to sell you makeup and CBD gummies, and try even harder to keep you engaged to see those ads, up until you set a school on fire. Cool. Great future we have.”

AI and gambling target the same people

“[…] thanks to all this money going into AI, despite nobody really knowing what to use it for, combined with the lack of A.I. regulation being something the Trump Administration brags about, it’s becoming a “Jurassic Park” situation if everybody had their own shoddy “Jurassic Park” in their pockets. But at least I know why we need a “Jurassic Park”. At least you get to see dinosaurs with a “Jurassic Park.” I don’t need a park where I get to see my dead grandma. We already have that, it’s called a cemetery. Anyway, this sucks, is my point. We all know it sucks. Why are we doing this thing that sucks? The only people who would want this are at rock bottom. Like “Timecop” levels of drinking in the dark and watching videos of your dead wife. Like I know it’s easy to say “wow that’s like ‘Black Mirror,’” but it’s literally an episode of “Black Mirror,” minus the freaky robot body. All this does is cheerily prey on the most fragile state of mind of people who either fear for or are grieving the loss of a loved one. It is designed to keep you from healing and moving on, for a subscription fee, by the way.”

Driving crazy people even crazier, faster

“According to research, lonely people are far more likely to anthropomorphize things. Of course we don’t need research to know this; just ask Wilson the volleyball that Tom Hanks definitely (beep) on that island. The actor, not his character. So you take this human trait and you add a product that specifically talks back to you in a way that agrees with everything you think, and you basically get a machine that catches people at their most vulnerable and feeds their worst impulses until they are removed from reality.”

Replacing friendships and therapy

“As it stands, a third of the people in the United States live in an area with a shortage of mental health professionals and even those with access likely never could or can no longer afford it. You combine that with a product that is unregulated to the point that it’s using emotionally manipulative tactics in order to prolong interactions, which, as mentioned, degrade more and more the longer you chat with them, that’s gonna be very bad!

“Heck, some chatbots are so desperate for your time and interaction that they’ll approach you first! Meta is training its A.I. chatbots to reach out to users unprompted and refer to past conversations to follow up on them. You know, like a friend. A needy, nosy, and manipulative friend who doesn’t care about you and just wants your money.

““Hey, Frank! How’s that divorce coming along? Did your son, Caleb, finally call? If not, maybe some Oreos, your favorite food, should make you feel better if you’re still too sad to masturbate. Also, your dog is spying on you.”

“It’s what happens when loneliness collides with unchecked capitalism. Instead of a country where mental health is provided to people and encouraged, we’ve built these busted ass-chatbots instead. And it’s gonna get worse. Because as I said, there’s no real need for these AI products for most people. The companies know this, but you bet your ass that they are reading the same statistics I am.

“And so, some tech ghouls are building LLMs specifically for therapy like Slingshot A.I., which has a chatbot named Ash that was designed and trained by psychologists, but isn’t actually a psychologist. Seems weird to name your therapist robot after the synthetic character in “Alien” who betrayed the humans and tried to choke Sigourney Weaver with a porn magazine for profit but whatever.”

A dependency machine

“See, see, see, there’s a fertility crisis and in order to increase birth rates we gotta, one, get rid of all the immigrants, preserve white culture, etc, but more importantly, to increase birth rates, we gotta get everybody hooked on fake girlfriends!

“Yeah, these people are garbage aliens. Of course they want you to use their dumb bots. For one, they make money if you do! But also, they seemingly have no idea how to interact with society without them. Sam Altman apparently doesn’t know how to raise his child without ChatGPT. Why would you use his product? He is literally saying that his product made him less able to function without it! You know, that cognitive debt we talked about!”

Scam your way to utopia

“I know I compared it to cigarettes already, but these are the tobacco CEOs talking about how great smoking is, and how they love to smoke, and then dying at 50, and not knowing why. And just like any addiction, this is a self-perpetuating problem. A crutch. Everything points to that. A person is lonely or shy and then turns to a chatbot to fix that, and the chatbot either keeps them hooked on their screens and makes them more lonely, or makes them unable to function without it until they can’t fucking talk to their child without consulting a machine, that hallucinates. It’s bad. And fuck. It’s like those fucking products you see in infomercials that offer solutions to problems nobody ever had. Except this particular SlapChop costs hundreds of trillions of dollars with no clear return. Let’s keep it that way!”

Stern robot wants to fix your life

The article The AI boom is causing shortages everywhere else (Hacker News) includes the following comment that attempts to not only put the capital expenditure in AI technology into context but also describes the immense opportunity cost.

“It’s hard to comprehend the scale of these investments. Comparing them to notable industrial projects, it’s almost unbelievable. Every week in 2026 Google will pay for the cost of a Burj Khalifa. Amazon for a Wembley Stadium.

“Facebook will spend a France-England tunnel every month.”

“ As a research topic, modern AI is a miracle, and I absolutely love learning about it. As an economic endeavor, it just feels insane. How many hospitals, roads, houses, machine shops, biomanufacturing facilities, parks, forests, laboratories, etc. could we build with the money we’re spending on pretraining models that we throw away next quarter?”

This is madness, of course, but it’s par for the course: some of us have observed the money machine pouring capital expenditure into military-hardware companies for a long time now, always to the detriment of any social investments, like those listed in the comment above: schools, daycares, public transport, “hospitals, roads, houses, machine shops, biomanufacturing facilities, parks, laboratories,” etc. That our elites are spending money on things beneficial to them rather than us is not new but that they’ve started wasting it on AI is.

Examining opportunity cost

Another commenter wrote that they’d,

“[…] just made a LLM recreate a decent approximation of the file system browser from the movie Hackers (similar to the SGI one from Jurassic park) in about 10 minutes. At work I’ve had it do useful features and bug fixes daily for a solid week.”

Implicit in this argument is that this person having recreated a file-system browser from a movie for pure fun—and with little to no effort on their part—did so in the context of a society that actually considered the costs, and billed the appropriate parties. That is, the commentator probably spent a few bucks on it but that’s only because society is massively subsidizing things that make the richest people richer and the richest people are currently obsessed with AI as the thing that will make them richer. People like the commentator are currently under the umbrella and benefit from the self-aggrandizing activity of the rich, at least tangentially.

For the rest of us, we really should be thinking about the opportunity cost, and we should very much be wondering why we’ve decided—mostly implicitly—that this person gets to have an infrastructure for playing around with their hobbies—or for being slightly more efficienthaving slightly more fun at work—while people in need of hospitals or medical care can go and hold a bake sale [1] or start a GoFundMe. This is fine, apparently, especially if we don’t really think about it.

LLM output is still mostly meh.

On top of it being an a moral affront that our societies prioritize military hardware and the self-aggrandizing dreams of tech elites and crypto bros above anything even remotely useful to everyone else, the output of these tools is still just not very good. They seem to have given up on making it better, and instead focused their efforts on media campaigns, advertising, press releases, and shills, all of which is meant to make us lower our standards instead.

The results may kind of suck, but you barely had to lift a finger to get them, and you can easily convince your boss—who also doesn’t care about the results—that you’re working while the little circle spins on your laptop, so it’s a win-win-win, right?

For example, here’s a picture illustrating the point of the previous section.

Military Expenditures vs. Social Spending

Do you like that graphic? Does it feel appropriate? Because, by way of illustration, I generated that image with Microsoft Copilot when I couldn’t find anything matching “military versus social spending scales” on DuckDuckGo that wasn’t a pie chart. I could have gone with a pie chart but I kind of wanted the scales. I’m spoiled too. I’m under the umbrella.

The result is kind of … off, though, isn’t it?

1. There is no physicality to it; it’s not grounded; it kind of floats.
2. The shadow doesn’t represent the two platters.
3. The rocket isn’t sitting on anything.
4. The wires holding the social-spending platter don’t go anywhere.

It’s fine, I guess? The results have gotten better but this is still where we’re at in 2026, after nearly $1T of capex and untold amounts of subsidies for energy infrastructure and data-centers. Given all that, that image is a wildly mediocre result.

It’s better than what I could have come up with in the minute that it took to generate it. But is it good? Absolutely not. It’s better than nothing. That is what we’re trading social services for: better-than-nothing versions of things we didn’t really need. So, elites (like myself) can control vast amounts of resources in the cloud to generate useless gewgaws while people go hungry and without healthcare.

This is the bargain. It always has been. [2]

Baubles for the rich always beat food for the poor

It’s different this time

Another comment on the Hacker News post writes,

“Is it the beginning of the star trek ship computer? If so, it is as big as the smartphone, the internet, or even the invention of the microchip. And then the investments make sense in a way.”

This is the one. Can’t miss.

Keep telling yourself that, buddy.

People keep claiming that these tools will eventually turn around and solve all of the other societal problems, which is why it’s absolutely sensible, patriotic, and moral to put all of our eggs in exactly this basket, this time.

Forget about all the other times. Lord knows we’ve tried hard enough to get you to forget about all of the other times we’ve scammed you.

Even if you vaguely remember that something unpleasant might have happened once or twice—or might even vaguely remember who was responsible—just forget all that. Because this time, it’s different. This time it will work. Promise. Cross our shriveled little hearts and hope to die.

Seriously, don’t even worry your pretty little head about it: There is no way this will turn out to enrich all of the usual suspects, leaving the rest of us with nothing. No way. This is the one. This time it’s real.

The same assholes who already own everything are recruiting you into their propaganda campaign to increase their fortunes. They’re saying, once again, “Let’s just do this thing first, then we’ll get to all of the things you need. Don’t worry, we won’t forget you.”

Lucy is soooo convincing, isn't she?

Maybe LLMs will kill LARPing I dunno

When I read about people building five projects a week, or submitting 27 PRs a day, I’m reminded of people who say that they read 200 books a year. This may be superficially true, but they are almost certainly crap books, or they’re just skimming them, or they’re incapable of understanding them. They are cheating. They are rounding up. They are emphasizing quantity over quality, which, like, used to be a bad thing.

Now, the barrier has been lowered even farther. People can now write 50-page “papers.” They can write “full-fledged apps.” Because the barrier to entry has been drastically lowered, there is less room for those LARPing as writers or programmers these days, not because they can’t LARP anymore but because no-one can tell the difference between their LARPing and LLM-supported LARPing.

A dozen years ago, the doors were wide open for people who could barely spell JavaScript—and had no idea what the difference was between that and Java—to earn six-figure salaries while building careers in an industry they had no hope of understanding. There was a lot of money sloshing around in the industry and managers greedily took up the slack in order to fill their teams with heartbeats.

These managers weren’t interested in actually accomplishing anything, but did it in order to look like they might accomplish something for long enough for them to get promoted like a space shuttle achieving orbit. They dropped their team like booster rockets, which careened back to Earth, only to be picked up by another enterprising manager more interested in a career than in actually accomplishing anything.

This worked out great for everyone as long as the industry was awash in money for such escapades. It no longer is, as those with all of the money have moved on to playing much larger games that don’t involve minor cogs earning six-figure salaries. They are instead focused on landing ten-figure deals that also have no hope of ever providing any value outside of making them money, but that’s the play these days apparently.

Long story shot, the LARPers are now having a tough time of it. They LARPed for so long that they think that they’re actually engineers whose jobs are endangered by the new fad on the block, LLMs. No-one cares about SPAs anymore because people who don’t care whether products actually work over the medium- or long-term can have any monkey churn out dozens of them a day to show their bosses. who also don’t care whether the products work because our entire economy is built on LARPing with only awards and no accountability or consequences in sight for the greatest transgressors.

Does it matter if anything even works anymore?

I think it’s premature to predict the end of anything when it’s completely unclear in what form any of what’s available today (A) will be available in that form and price point in the near future and (B) whether it even is what it claims to be—or what its most fervent acolytes claim it to be.

Hype is hype because it grows by repetition rather than by the introduction of new information. We are seeing a giant version of that and it feels inevitable because a lot of people are spending a lot of money to make it feel that way.

if you know what you’re doing, then you personally should have nothing to fear because you and I both know that the future will not be herding LLMs because it doesn’t work the way they say it works, nor will it until something significantly changes.

Since no-one seems to be interested in going anywhere near a drawing board to do some basic research, and since the amount of money being sloshed around to support the current LLM-based fantasy is larger than anything we’ve seen before, the aftermath is going to be epically bad.

I think that we can safely say that losing our jobs to AI will be the least of our concerns as we pick our way through the pillaged aisles of an abandoned grocery store in the post-apocalyptic hellscape that is definitely coming in the next financial crash that will make 2008 look like a bank error in their favor. [3]

Community chest bank error in your favor

Show me what you can do now

The verb case the proponents of this revolution use is always “future”. Success is just over the horizon. Just a little bit more. This is how MLMs work; it is not a serious business model.

These people love to round up. “We built a browser”. STFU. You did not. You built another prototype. You built another copy of something that already exists.

We are over four years into this mess and all I can see is software getting noticeably worse.

Hey, you jumped your bike over the ramp again, Billy. Cool. Can you do something useful yet? Like, can you go to the store and get me some goddamned cigarettes?

They have every incentive to lie bigly

All I see are fantasies spun by people worth hundreds of billions of dollars who run companies that are hundreds of billions of dollars in debt who are trying to keep the plates spinning so that you don’t notice that some of these people are planning to leave by the back door real soon.

When the CEO of Anthropic tells you that his company is going to change the entire world, it’s the same thing as when Trump says that polls no longer matter. They desperately need you to believe these things even though they don’t believe it themselves.

I think a great example of this is when Tesla quietly abandoned its autopilot program a little while ago, after years and years and years of telling people that they would be able to drive their own cars without touching the wheel—and after several people actually believed it so hard that they killed themselves in car accidents. Now, years later, that program is just completely gone. It is no longer officially a program just like it was never an actual non-imaginary thing to begin with.

That’s all a lot of this is. I’m happy if you find value in it for yourself. But it’s not here to provide value to you. It’s here to pretend to be much more valuable than it is so that a bunch of rich people can cash in, get out, and move on to the next scam.

I would have thought that a lot more people would have grokked that none of the people claiming that they’re here to save the world for all of us are even slightly interested in doing so, to say nothing of being capable of pushing us in that direction, if only a little bit.

Our radar for detecting grifters is completely broken, and that’s why society is saturated with them, top to bottom.

The article AI Doesn’t Reduce Work—It Intensifies It by Simon Willison demonstrates the Overton Window of addiction pretty well. The author writes,

“This captures an effect I’ve been observing in my own work with LLMs: the productivity boost these things can provide is exhausting.

“[…] I’m frequently finding myself with work on two or three projects running parallel. I can get so much done, but after just an hour or two my mental energy for the day feels almost entirely depleted.”

Is it marketing?

If I didn’t trust the author, I would immediately suspect that he’s applying a classic marketing tactic: “OMG this tool is so powerful that I can’t even control it!” [1]

This is a classic tactic of trying to sell a product by arguing that you probably shouldn’t use it, not because it doesn’t work, but because you probably couldn’t handle it. This is the kind of marketing that appeals to children, teenagers, and Jackass fans.

Instead, I think he might be seriously not noticing that his argument amounts to, “All of this cocaine I’m doing has doubled my productivity but I can only work a quarter of the day. Also, I feel like shit.”

Maybe you’re using it wrong?

A snarky response would be: “Hey! Here’s an idea I’ve heard somewhere: maybe you’re not prompting it correctly.” [2]

But I’m not going to be snarky.

You’re not special

The author continues to describe what he clearly seems to think is a unique phenomenon. It’s only unique when you’re trapped in an information bubble where you start to attribute every detail of existence to the thing that you have grown to unreservedly love.

“I’ve had conversations with people recently who are losing sleep because they’re finding building yet another feature with “just one more prompt” irresistible.”

My friend, you are describing addictive behavior.

It was no different before LLMs. This is how it has always been with programming (or for any interesting task, really). I spent a lot of my early 20s programming day and night.

As you get older, though, you learn that just leaving it be is just as efficient. That is, instead of staying up two or three more hours growing increasingly frustrated, you can finish whatever you’re working on in the morning—and probably in five minutes. That’s almost always the more efficient and sustainable solution.

But, sure, let’s pretend that this behavior is unique not only to programming, but to programming with LLMs.

“I think we’ve just disrupted decades of existing intuition about sustainable working practices. It’s going to take a while and some discipline to find a good new balance.”

“We’re all doing too much cocaine, right guys?”

Sure we are, Simon. Sure we are.

The following video is an excellent analysis of the state of AI-generated content as used to generate false narratives that are politically advantageous to the elites. Evan focuses on accounts and influencers that promote the narrative of an increasingly lawless and violent London that use completely fictitious, AI-generated content and which benefit personally tremendously from the advertisements shown on their “engaging” content.

Something Very Strange Is Happening To London by Evan Edinger (YouTube)

The locations in these extremely popular videos that he examines either don’t exist or they’re in towns that are nowhere near London. Many of the posters are probably not even real people or real accounts. They peddle lies to generate anger, then harvest attention, funneling it to advertisements. Evan argues that the monetization on videos like this should be disabled immediately. It’s a good idea but it will never happen. He further recommends to get outside, to experience life in the city to see that there’s no truth to anything that you’re seeing online.

Violence in the U.S.The tactics he covers are the same as those used to manipulate public opinion about the violence in any of a dozen U.S.-American cities. None of the violence purported to exist by the administration actually exists, but the Trump administration used it as an excuse to send national troops to several cities that he was otherwise mad at. The president continues to refer to this nonexistent violence to this day. None of it this stuff is real but it has real-world consequences.

The article John Haugeland on the failure of micro-worlds by Mark Dominus (The Universe of Discourse) describes a very early experiment in natural-language processing from the late 1960s called SHRDLU (Wikipedia).

“SHRDLU could handle this too, although I think its mechanism was different: it would interact with the separate blocks world subsystem and ⸢actually⸣ try to put the block on the pyramid; the simulated physics would simulate the block falling off the pyramid, and SHRDLU would discover that its stacking attempt had been unsuccessful. With Claude, something very different is happening; there is no physics simulation separate from Claude. I think the answer here demonstrates that Claude’s own model includes something about pyramids and something about physics”

Does it though? Where would it have acquired this model? Why would it suddenly be modeling physical laws unless some layers surrounding the text generator had been bolted on? As an engineer, I would love to know how much of what goes into these answer is actually located somewhere in calculation units that have nothing to do with a transformer-based, attention-enhanced LLM.

If it’s the LLM doing it, then I don’t know which part of its architecture the answer comes from. I don’t see the mechanism because, so far, we’ve managed to explain a tremendous amount of its “behavior” (responses) with statistics. Is there a reason to have stopped assuming that this is the mechanism underlying the supposedly improved “understanding”? What is the reason we’re rounding up now as opposed to two years ago?

As far as I know, we’re just throwing more horsepower at these tools but haven’t significantly changed the architecture that would lead us to believe that a “world model” is now governing token choice rather than “statistics”. I might have missed something, though; my attention wanders.

“Are there are any people who are still saying “it’s not artifical intelligence, it’s just a Large Language Model”. I suppose probably.”

Well buddy, I don’t spend any time talking to these things, so I admit that my thinking kind of got stuck at that stage. I think that other people starting rounding up to “this is intelligence” because they started having too much fun with it and they didn’t want it to look like they were just playing video games.

So, instead of talking about the mechanisms that go into these models—if they’re at all different from what we presented a few years back—they talk about how it “seems intelligent” … and how everyone who doesn’t think this is a benighted fool, if not a heretic.

I can’t help but recognize that this mechanism is the same as the one employed by any true believer in any other faith. First, you’re “born again”. And then you start to castigate anyone who isn’t. Classic MLM tactics. Human psychology is utterly banal.

“But as a “Large Language Model”, Claude necessarily includes a model of the world in general, something that has long been recognized as an enormous prerequisite for artificial intelligence. Five years ago a general world model was science fiction. Now we have something that can plausibly be considered an example.”

There it is again: the author is “rounding up” quite significantly because he doesn’t have another explanation for what he thinks he’s seeing. Does the LLM have a model of the world encoded within its statistical matrices? That’s quite a claim, seemingly belied by the many, many times that it gets things wildly wrong. Is it that it has a model of the world but is kind of dumb sometimes, like a child?

If it had a model of the world, then why is context so important to keep it on the primrose path? What is the theory here? Is it that the author wants it desperately to be more than it is? Would he marry it? Invite it to dinner? Watch a movie with it? Maybe.

“And second: maybe this isn’t “artifical intelligence” (whatever that means) and maybe it is. But it does the things I wanted artificial intelligence to do”

You’ve found a tool. You’re happy with its functionality. Good for you. I have completely different expectations and quickly grow bored because there are only so many hours in a day and I am not in any way attracted to spending any of them talking to a chatbot.

There is too much of actual human culture and art out there for me to bother with artificially generated content. Was there too little of everything before? Did we not have enough books or movies? Are there not enough people to talk to? What are we even talking about here?

The article macOS Tidbits by Jasper Lai has dozens of tips but I’ve only included the ones below that I had either never heard of or that I’d forgotten. There are still a lot of them.

1. “⌥ + ⌘-click an app in the Dock to switch to that app and hide all other apps at the same time. This is great when screen sharing.

   “Hold ⌘ to interact with background windows without bringing them into focus.”

2. “[…] double-click and drag to select word-by-word. Triple-click and drag to select paragraph-by-paragraph.”

3. “When taking screenshots, hold ⌃ to copy the image instead saving it to your desktop.

   “When using ⇧ + ⌘ + 4 to take screenshots, press space to capture by window. In this mode, you can also:”

   • hold ⌥ to take the window screenshot sans-shadow; and/or
   • hold ⌘ to capture child views within a window (such as New/Open/Save dialogues, alert windows, et al).

4. “Any self-respecting Mac app opens the Help menu when you press ⌘ + ?.”

5. “Hold ⇧ + ⌥ to adjust display brightness, volume or keyboard brightness in quarter-increments. This is useful when the lowest click is still too bright or loud.

   “A quick way to access your Displays settings is to ⌥-press either brightness up or brightness down.

   “Same goes for Sound settings: ⌥-press mute or volume up/down.
   Again with Keyboard settings: ⌥-keyboard brightness up/down.
   (Works with Touch Bar too! ⌥-tap the corresponding button in the Control Strip.)”

6. “In Finder, hold ⌥ to Get Info on all selected items in one Inspector window, rather than in a barrage of individual Info windows. This also works with ⌥ + ⌘ + I< (instead of ⌘ + I).”

7. “You may already know about the Go to Folder… menu item (⇧ + ⌘ + G) in a normal Finder window. This is even quicker to invoke from an New/Open/Save dialogue: just hit /. (The usual shortcut still works.)”

8. “With any standard column view (such as in Finder), hold ⌥ to resize all columns equally.”

9. “⌃ + ⏎ to right-click whatever is currently focused. (Though, strictly speaking, there’s no clicking involved here.)”

   I have been looking for this for years … but it doesn’t work. However, it inspired me to finally figure out how to do trigger the secondary mouse action with the keyboard.

   1. Open Accessibility => Pointer Control
   2. Check the box for Enable alternative pointer actions
   3. Select Options…
   4. Choose the keyboard combination that you want.
   5. I assigned ⇧ + F10 to match my muscle memory from Windows.

10. “⌘-click items in the Dock to reveal them in Finder.”

The article Prompt caching: 10x cheaper LLM tokens, but how? by Sam Rose (ngrok) included the following hypothesis,

“[…] what if we had a problem where we didn’t know the formula? What if we just had this mysterious table of inputs and outputs below?”

Table of inputs and outputs

The author wrote,

“I will say that ChatGPT figures it out straight away if you paste a screenshot into the app.”

Holy shit! Really?

I opened up https://chatgpt.com for probably the first time in my life and pasted the screenshot and asked, “What function produces this output” (I used “What” and no question mark so that ChatGPT might think I’m a cool Get-Z-er instead of a cynical Get-X-er).

Nothing up my sleeves − the entirety of my prompt

It thought for 30 seconds—though at least half of that time seems to have been running OCR on the image—and produced this absolute masterpiece.

Isn’t it beautiful?
Do you see how nice the formula looks?
Do you see how it worked out each of the values?
Do you see the little check marks to indicate that it got the right answer for each and every one of them?

Breathtaking.

Do you see the confidence exuded by the emoji ✅ followed by “This function matches every row in the table exactly.”

Go big or go home.

ChatGPT's answer, after thinking for 30 seconds

Before I had scrolled below the fold to see the examples, I had already mentally started popping values into its formula for the first line in the table and had come up with 67 instead of 73 but apparently I can’t math because look, there it is in ChatGPT’s answer: 22 = 10. Q.E.D.

It’s funny that it managed to sort the input values, even though that’s a very confusing way of showing a proof for a table of values that was not sorted.

Look at that beautiful formatting, though.

4 + 1 = 3. Majestic.

10 + 4 = 29. Literal tears of joy.

1648 + 9 = 1277 Who needs a second coming when I can slip the surly bonds of Earth and dance the skies on laughter-silvered wings to reach out and touch the face of ChatGPT? [1]

I guess it still doesn’t work for me like it seems to work for everyone else. [2]

The article Smartphone Buyers Care Even Less About AI Than They Did Last Year, CNET Survey Finds by Abrar Al-Heeti (CNet) contains the following illuminating graphic.

Almost no-one cares about AI on their phone

“In 2024, the biggest motivation for US smartphone owners to upgrade their devices was longer battery life (61%), followed by more storage (46%) and better camera features (38%). Just 18% said their main motivator was AI integrations. This year, it appears that number is even lower, even as AI capabilities become more ubiquitous. ”

“Just 13% of people say they use AI on their phone to summarize or write text, 8% say they tap into AI image creation tools and 7% use AI on their phone for photo editing. Additionally, 20% admit to not even knowing how to use the AI features on their handset.”

That’s not surprising and it’s probably not just the AI feature, so beware of this statistic. These are people who barely know how to use anything on their phones. They use any device by ritual. If an icon moves or changes color, they’re lost

On the other hand, the low-usage numbers are damning. People aren’t using AI features and don’t care that they might be missing out on something. In a world of FOMO, and with the incredible push for AI, this is really damning.

It may very well be that the hype is hyper-focused on the tech world and the rest of the world doesn’t even really notice this stuff. It may be that they’re finally really trying to sell something too complicated for people to use or even know what to do with.

“Samsung, for one, says on its website that Galaxy AI features “will be provided for free until the end of 2025 on supported Samsung Galaxy devices.” Apple is also expected to eventually start charging for some of its AI-powered iPhone features. You’ll also need to pay to unlock Gemini’s full power across Google’s apps. Amid so much subscription fatigue, that could be a tough sell. Half of people surveyed say they’re not willing to pay extra money to access AI features on their phone. That’s up 5% over last year.”

Wait. Almost no-one is using AI features but only half of all users would be willing to pay for those features? That implies that there is a large subset (1/3?) who would be willing to pay extra for features that they don’t use. Oh, never mind. That tracks.

Actually, according to the numbers in the chart below, only 3% of all adult users are willing to pay for AI features, and 50% said that they would expressly not pay more.

Users don't know how to use AI, don't want more, and don't want to pay for it

I can’t help but include the methodology section at the end of the article because it was so cool that they included it in such detail.

“CNET commissioned YouGov Plc. to conduct the survey. All figures, unless otherwise stated, are from YouGov Plc. The total sample size was 2,201 adults, of whom 2,129 own a smartphone. Fieldwork was undertaken May 13 to 15, 2025. The survey was carried out online. The figures have been weighted and are representative of all US adults (aged 18 plus).”

The article Where’s the Shovelware? Why AI Coding Claims Don’t Add Up by Mike Judge is an interesting read that makes the following argument, more or less,

“If so many developers are so extraordinarily productive using these tools, where is the flood of shovelware? We should be seeing apps of all shapes and sizes, video games, new websites, mobile apps, software-as-a-service apps — we should be drowning in choice. We should be in the middle of an indie software revolution. We should be seeing 10,000 Tetris clones on Steam.”

As bad as the world of existing non-LLM-generated slop already is, the author’s point is that it’s not gotten measurably worse.

If AI allowed pretty much anyone to build an app—the proposal buoying the AI bubble—then we’d be flooded with a tsunami of crapware rather than just drowning in a ocean of it.

“[…] billions of dollars have been invested in these tools. Billions of dollars will continue to be invested in these tools. The problem is that they’re being sold and decisions are being made about them — which affect real people’s lives — as if they work today. Don’t parrot that nonsense to me that it’s a work in progress. It’s September 2025, and we’ve had these tools for years now, and they still suck. Someday, maybe they won’t suck, but we’d better see objective proof of them having an impact on actually shipping things on the large.”

From the comments on Reddit:

“Today (actually not joking) a manager told me”

“AI should make you 10x more productive, what takes you 10 days should take you 1.”

“Which I figured was bullshit because Tuesday he asked”

“Can we compile OpenSSL v3.6 for RHEL-5? Docker makes this easy right?”

“IDK how AI makes me 10x more productive when I spent 4 hours in meetings to realize we actually needed to update our LuaJIT (on RHEL-10) not compile a version of OpenSSL (???)”

Streetlight effectThis is a much better point to consider. People are searching for their keys on the sidewalk under the streetlamp when they lost them in the bushes. Getting people to address inefficiencies in priority order would be a much bigger lever than letting them take the easy way out by bike-shedding with AI or by trying to force people to USE AI DAMMIT to run in the wrong direction.

What’s the point of doing something faster when it doesn’t need to be done in the first place?

This Record Label Is Trying To SILENCE Me by Rick Beato (YouTube)

Rick Beato was forced to hire a lawyer to defend his fair-use playing of artist’s music in his videos. The labels abuse the copyright-strike system and Google cheerfully goes along with it.

He has “successfully fought thousands of them—never lost one—they still keep coming in.”

There is no way for him to defend himself against these without a lawyer. UMG (Universal Music Group)—or, most likely, the third-party firm that they hired to enforce their copyrights—are not punished at all for raising invalid claims against people who are rightfully claiming fair use.

If they’ve failed at thousands of claims, why should they get to continue to lodge complaints for free, wasting everyone’s time, and making it more difficult to create interesting interviews and analysis? Google clearly doesn’t care, as this has been going on since nearly the beginning of their purchase of YouTube.

This is the world they have built for us. They hate us. They despise it when we do anything that doesn’t make them money.

Get back to work, monkey.

This 21:36-long video is chock-full of useful information: use a real VPN (not a free one; be sure of the vendor), hide your real email address wherever possible, stop clicking sponsored links in search results—although he doesn’t recommend to use a search engine other than Google—, use an authenticator app for 2FA instead of text messages, etc.

The Truth About Those Age Verification Pop-Ups by Evan Edinger (YouTube)

0:54  Details of the UK’s Online Safety Act
3:19  Recent “unavoidable” Data Leaks 
4:55  Why the Online Safety Act Immediately Fails
7:10  How Free VPNs can decrease your data privacy
8:24  How the Online Safety Act is filtering the news
9:10  How the UK Looks on the World Stage in Technology
10:30 How little Parliament seems to know about VPNs
14:25 How to actually keep your data private online
15:16 My best tip for searching Google
17:13 Don’t set your 2 factor authentification up wrong
18:09 How an Internet Router and VPN Work
20:31 How the UK’s Online Safety Act will affect UK businesses

He also doesn’t mention using a password manager, which is like the thing you should do. I can’t recommend passkeys yet because I haven’t started using them yet but I probably could easily do it with ProtonPass.

Will this advice keep you safe? The title of my article says “more safely.” There is no guarantee. But it’s better than using the the same password everywhere and clicking on everything in sight.

Building awareness helps. The other day, I was logging in to an issue-tracker for a vendor and noticed that my password manager wasn’t offering to help me log in. It was because I wasn’t on the vendor web-site anymore. The URL was being redirected to another URL. The new URL looked plausible but it was a different FQDN (Fully Qualified Domain Name). So now what?

Now what? Now you don’t log in.

Instead, I wrote to the vendor and they immediately responded to apologize for the inconvenience: they’re having trouble with their own domain name, so they had to use the redirect for now.

In this scammy world of scams, it is best to exercise an overabundance of caution. The thing that you think you want to do isn’t important enough for you to give up everything else.

Scammers like to instill time-pressure and panic. Don’t give in to it. Take a breath. Think about it. Verify through another channel.

I, for one, welcome our new AI overlords.A friend sent me the article Finger weg von den neuen KI-Browsern by Michael Andai (20min) (“Hands off of the new AI-browsers”).

The article largely focuses on the grievous security holes in these browsers, making them not browsers but data-exfiltration apps. In an age of unprecedented scammery, it is an affront that these tools even exist.

But that’s not even the worst of it.

With a web browser, you type in an address and see the content hosted for that address. You trust your browser to deliver—unfiltered and unchanged—what you asked for. This implicit trust is extremely important, as the data your browser returns informs your worldview. [1]

These aren’t web browsers. They don’t find content; they produce content. You don’t actually see any web pages themselves when you “browse” with these tools. Instead, you see summaries generated on-the-fly that serve as a “response” for your “request”.

To be clear: you type in a prompt and see what the LLM generated as a response for that prompt. I would imagine that a lot of the pictures and short videos included in these responses are also generated. You will not see anything that anyone actually produced, unfiltered. You are implicitly trusting that tool—and the company that produces it as well as the laws of the country where that tool’s infrastructure “lives”—to deliver a reliable worldview.

For those of who use the web without an algorithmic feed, this feels like a significant change. It feels like madness to even think of using a tool like this. For people who have already been trained to simply look at what they’re shown, this is more of an increase in the level of control that platforms have already had over what their users see and hear. They’ve been trained to not give it a second thought.

Although it’s not technically a significant difference over what a Facebook, TikTok, or Instagram feed already did, it is a big step in the wrong direction down a road these people shouldn’t even have been on in the first place.

If you can train people to become accustomed to this, then it opens the door to further great leaps forward for controlling what they see and hear.

The level of control over what people see and hear is already too high for comfort. However, while your ability to directly access content is sometimes impeded with a web browser, it has, until now, never been transformed or interpreted.

This is very, very different.

LinkedIn is an enshittified dumpster fire.

I recently set up 2FA for my LinkedIn account. Then I changed the email address associated with that account because the old one was an ancient throwaway that I’m phasing out. Not long after, LinkedIn blocked my account, ordering me to provide a real-world physical proof of identity in order to get back in to my account.

LinkedIn has blocked my account as a precaution

“We take proactive steps to secure your account when we detect potential unauthorized access. Signs that your account may have been compromised
include account access from unfamiliar locations or devices, or unusual activities such as multiple attempts to change passwords or critical settings.”

I was trying to make my account more secure and they considered that suspicious. Even weeks later, after no-one has done anything further with the account, they’re not freeing it back up. So my account on LinkedIn is stranded, encysted in the past. No big loss, really. It was very useful when I was looking for a job; I honestly hope I won’t need it for that again.

“To help you keep your account safe and regain access, we’ll first need to confirm you’re the valid account owner. Please submit a government-issued
ID to start the process.”

I haven’t been on LinkedIn for weeks now because I am not giving a social-media site a picture of a government-issued identity.

“A customer support representative will contact you within 48 hours with more information about your account and the status of restoring your access.”

This has never happened. They certainly didn’t send a mail to the account associated with my LinkedIn account. Below you can see the most recent messages from LinkedIn to that account.

Successfully configured 2FA and a new email address for LinkedIn

I tried again this morning and was still blocked in the same way.

So, I checked what my options are. The Verify your identity to recover account access page gives several options, each of which involves the LinkedIn organization not being satisfied with my password and not even trying 2FA (you know, to see if I would be able to respond to that) because they apparently don’t trust their own security mechanisms.

Your options for recovery are:

• Recover your account using a government issued ID
• Verify your identity with something called “Persona”, which I’m sure is a third-party identity-management company that totally doesn’t sell your data to all comers.
• Other ways to recover your account

  • Through an Affidavit of Identity “sign[ed] before a Notary Public” (like, who the f@&k do you think you are?)
  • Through a work email address

I would be interested in the last option but no-one has contacted me and that option is not available through the web site—probably because most people would just use that option instead of uploading a picture of their most precious form of identity to a social-media site with delusions of grandeur.

So, I went to Contact LinkedIn customer support

Step one to get support for not being able to log in: LOG IN

Near the bottom of the page, they have a “tip” that reads “If your account is blocked or restricted, learn more about account restrictions.”.

Here's a tip − go f@&k yourself

That link takes me back to a page that tells me that the only way to reinstate an account that has been blocked for “identity violations” is to verify your identity as outlined above.

“Our Professional Community Policies require members to use their true identity on LinkedIn, provide accurate information about themselves or their organization, and only share information that is real and authentic. If we find that your profile or parts of your profile are intentionally fraudulent or do not reflect your true identity, we may place a restriction on your account. Depending on the type of violation, access to your account may be restricted either temporarily or indefinitely. If you believe your account was restricted in error, please login and follow the onscreen steps to verify your identity.”

OMG really? Because then you’d have to ban most of your users because everyone lies their face off on LinkedIn all of the time.

“Additionally, if we find signs that your account has been compromised or taken over by another person or entity, we may take proactive measures to restrict your account to protect your information. To regain access to your account, please login and follow the steps on screen to verify your identity.”

I guess this is what happened but it’s only because LinkedIn has implemented a least-common-denominator approach to support and moderation because they know that they can just cheerily demand that their users do a bunch of extra work [1] in order to partake in their vibrant community of liars and bots.

No. You guys f@&cked up. I have a list of mails above that show how happily you allowed me to set up 2FA and change and confirm my email address. If you consider that to be suspicious activity then kiss my ass.

I’m still waiting for that promised contact from LinkedIn. Until then, I guess I’ll have to live without that flood of increasingly AI-generated content.

The article Function calling using LLMs by Kiran Prakash describes an approach that works very well when you don’t have a testing environment: build a plan, evaluate validity of the plan, and then apply the plan after verification. You should also be able to slice the work into sub-tasks to make verification more reliable.

This is the approach I took for a PowerShell script that runs against ADOS (Azure DevOps): it’s production data, so you really want to be sure what is going to be executed, but you have no testing environment or ability to get one.

“It’s important to emphasize that when using function calling, the LLM itself does not execute the function. Instead, it identifies the appropriate function, gathers all required parameters, and provides the information in a structured JSON format. This JSON output can then be easily deserialized into a function call in Python (or any other programming language) and executed within the program’s runtime environment.”

In the implementation, you can see how the code he writes prepares the query to the LLM in a structured way, providing the context in an attempt to guide the result. Happily, he begins by writing unit tests!

This is another good step-by-step example of working with an LLM, but for a different task: it’s using an LLM as an interpreter for the user’s input. It’s basically a way of adding a natural-language “search-like” interface to an app without forcing the user to structure their input, without developing an UI, and without writing a parser. The advantage is that you get a way of querying a potentially large API surface in a way that in more amenable to more users.

I think of an example from Markus Schenkel from Cudos, who talked about using an MCP plugin for working with a CAD/CAM program—apps that notoriously have dozens of toolbars and thousands of functions. He could formulate his “novice” request as text, and the LLM, together with the mapping to tool functionality, made relatively good guesses about what he was trying to do. It often took a few attempts—but he was able to accomplish his task, whereas he would have either given up or had to invest a lot more time to get it done otherwise.

I think this is great for products that are in proof-of-concept stage, so that you don’t iterate on UIs too early in the design process. But we also have to be aware that we have UIs for a reason. Once there’s a well-established set of use cases and functionality, then it’s unclear that making users continue to use a command-line interface where they compose text is better than a GUI.

At any rate, the article is filled with detail and code (in Python) for using an LLM in the way described above. There’s a section on refactoring at the end, a comparison to the rules-engine-based approach that this technique seeks to replace, and also a comparison of function-calling with MCP.

Chana MessingerThis ~10-minute video discusses research about chain-of-thought LLMs that “show their work”. Chana points out that, once you can see what the machine says it’s doing, it’s actually openly discussing “cheating” to achieve the correct result. She says that, once you add penalties for “cheating”, the machine doesn’t stop cheating—it simply stops writing about it. While this feels hilarious because it really seems to be acting like a teenager, it’s exactly this kind of anthropomorphizing that is both so seductive and also potentially counterproductive.

'Forbidden' AI Technique − Computerphile by Computerphile / Chana Messinger (YouTube)

Anthropic published a long paper recently called Circuit Tracing: Revealing Computational Graphs in Language Models in which they note that their research shows that the explanation offered by an LLM for how it arrived at an answer does not always—or even often—correspond to the actual path that the solution-generation took through the model’s layers, when examined in detail with profiling.

Even though Chana says that the LLM is describing how it’s going to “cheat” at getting to the answer that it knows has the greatest “weight”—i.e., it’s the thing that the questioner very clearly wants to hear, or gets statistically closest to the “answer” that was given in the eval included in the query—it’s actually describing this in a part of its processing that is only associated with generating the chain of thought and has little to nothing to do with producing the actual answer itself.

What we consider to be the “chain of thought” is, to the LLM, just more text to generate. It’s just as likely to be completely made-up and has little to nothing to do with the construction of the answer itself. The LLM doesn’t “know” that it’s explaining one part of a text with another, just like it doesn’t “know” that it’s “lying” or “cheating”.

The LLM is generating an answer that best satisfies the weights in its model (generated during training), combined with the “pressures” included in the system prompt and the query. It’s the human interlocutor who imbues the situation with humanity or intent, not the machine. The context is that you’re “talking to something” and the interpretive gloss is wholly one-sided. The other side is just cheerily crunching numbers.

I’m not convinced by Chana’s explanation that the LLM is actually “hiding private messages to itself” with steganography because I think that the better explanation comes from the Anthropic paper linked above, not the OpenAI one she discusses. However, I think that it’s definitely good advice to avoid these types of validation pressures, not because the models are “trying to trick us, or hack us” but because they don’t lead to the desired result.

I think this research is fascinating because, even though there is no-one on the other side (or it’s one of Searle’s Chinese Rooms), we still might be able to figure out how to manipulate the machine to give us what we want reliably. While I understand that the anthropomorphizing explanation is more approachable, I’m more wary than many others of the limiting effect it has on how we think about solutions.

I ran into a small problem while upgrading Visual Studio 2022 to 17.14.0, so I reported it (Visual Studio Developer Community) with the following text,

“The Microsoft.VisualStudio.Community.MSI component could not be installed. At one point, the installer told me that another installer was running, but there wasn’t any installer running. An installer had run before the Visual Studio upgrade: JetBrains Rider. It’s possible that this interfered?

“I am unsure how this problem will affect my work. I don’t really use the MSI tools in Visual Studio (that I’m aware).”

I received a response relatively quickly, as follows. I have highlighted what I consider to be problematic passages that I addressed in my

“After reviewing the error you reported regarding the error with a VS Installer package.

“Please make sure you have disabled any antivirus, group policies or firewall that you may have on your machine as they sometimes avoid the installations. If they are disables and the error persists, please try the following:

“Step 1: Please go to this path: C:\ProgramData\Microsoft\VisualStudio\Packages

“Step 2: In the above path there should be a folder with the name Microsoft.VisualStudio.Community.Msi;PackageAction. Inside this folder there should be either a .msi or an .exe file, if there is and run it, if there is no .msi or .exe, please delete the folder.

“Step 3: Then go to the VS installer and if there is a “More > Repair” option, select that. If it only shows the option “retry” please select that.

“IT Crowd − Have you tried turning it off and on again?Step 4: If that workaround was not successful then, try to uninstall Visual Studio using the install cleanup tool
See: Remove all with InstallCleanup.exe. (After you run the command form CMD, please delete the “Installer” folder from the following path and retry the installation: C:\Program Files (x86)\Microsoft Visual Studio)

“Please let me know if the solution worked for you! If not, we will continue investigating your issue.

“Let us know if there’s anything else I can help you with.”

I replied as follows:

Thanks for you help. I’ve found the folder you described and run the installer.

I also found the More => Repair option but elected not to execute it because it warns me that,

“Repairing will reset the environment. Local customizations like per-user extensions and your user settings will be removed. Your synchronized settings will be restored.”

I’m not so interested in getting rid of that ⚠️ icon in the VS installer that I’m going to take the time to re-install all of my extensions. I’ll just wait for the next VS update to (hopefully) clean things up for me.

For the same reason, I’m not going to reinstall VS unless something I actually use has stopped working.

On a final note, I was somewhat surprised to see this advice:

“Please make sure you have disabled any antivirus, group policies or firewall that you may have on your machine as they sometimes avoid the installations.”

While I understand that Windows Defender can issue false positives, I’m not a fan of the advice “don’t worry bro, our installer will work just fine once you disable all of the security on your machine.” I mean, that sounds like something an actual scammer would say. No offense.

On top of that, there are a lot of users (myself included) who work on machines configured by other organizations who do not have control over antivirus or firewall on our machines. I don’t know what you mean by “disable group policies” because they are legion and cannot “all” be disabled.

Thanks again for the detailed instructions. They worked as advertised. If I run into more problems, then I’ll have to try the “Repair” option.

This podcast episode Who’s Coding Now? AI and the Future of Software Development by AI + a16z (Apple Podcasts) was recommended to me by a colleague. These are my notes that I took (and later cleaned up) from listening to this single episode.

Send a check or money-order to…

Near the beginning, one of the hosts says,

AI Bubble

“There was a good blog debate about whether we’re overinvested in AI. I think the number was $200B annual investment. And I think the question was how we would recuperate it?

“Well, here we have a way to recuperate $3T, which makes the $200B look like peanuts.”

Sure, sure … except that people have to invest $200B first and the guy is saying that a $3T market will magically appear. it’s not even close to that after three years of the biggest brains in the world working intensely and ceaselessly on it. There is no evidence for that market yet, but everybody’s saying that it’s definitely coming. This is called an ⁠echo chamber (Wikipedia) and it’s the perfect place to brew up market bubbles. The nice thing for these people—A16Z—is that, even if the $3T never shows up, they’ll still have gotten the $200B.

Programming is hard, bro

A little bit later, they’re discussing how they use the tools but they don’t talk about which problems they’re solving. One person said that they start with specs, which is great (seriously). The others talk about how “no-one can remember all of the CSS classes like margin or padding…”, which makes my eye twitch. It’s like hearing your car mechanic say, right before they’re leaning in to fix your car, “no-one knows what all these wires are for…”

The host sounds like it’s an AI reading pre-canned text. I don’t think that it’s a person in the conversation. It basically throws up straw-man, leading questions, like,

Haters gonna hate, yo

“Is there some way to get the neckbeards engaged?”

Ah, yes, if people don’t jump on board with your scam—or they threaten to try to dissuade people from getting suckered themselves—then disparage those critics as nerds, training your minions to be unquestioning monkeys who don’t want to be called names. Don’t you want to be a cool-kid, AI-tool user making tons of money? Or would you rather be a neckbeard/hater/loser who’s going to lose his job to the cool kids?

If it’s such an obviously good thing, then why do you have to try so hard to sell it? Is it because you’re selling a solution to a problem that people don’t know they have? Is the problem that they don’t have a problem that your tool can solve? Or that they don’t recognize they that have a problem? Why can’t the tool’s performance speak for itself? Why does it need so much hype?

You’re not using it right

A bit later, the lady Yoko Li says,

“Given enough context and given enough tools…”

The problem, as far as Yoko is concerned, is that people aren’t able to use the tools enough yet, otherwise they’d be even better at helping you! And maybe you need to spend $200/month to get it working…and if it still doesn’t work, then it’s your fault.

They very lightly discuss context-poisoning and how the models will cheerfully offer wrong answers rather than admit when they don’t know something. They don’t offer any advice about what to do about it (e.g., resetting context in order to resolve poisoning, but that’s a “nuke it from orbit” solution that may throw out the baby with the bathwater). One of the guys says that LLMs are really good at more-complex tasks, which I think he misspoke, but I can’t be sure.

They admit that “models are not really creative…” and then say that if you’re doing something new, then it won’t help at all. I think that’s actually wrong! They can still be used as code-completion, even if it would be useless to try to have the LLM design the whole thing (which kind of works for tasks that have been done a million times before).

History never happened

One problem I have with these kinds of podcasts is that they sometimes feel so outside of history and prior work. The people seem to be considering problems of how we learn, how we create, and other questions of philosophy for the first time, which makes their analysis pretty superficial—because they’re retreading territory that many others have already covered, sometimes for centuries, if not millennia. I find myself thinking, yeah, that’s Kant, yup, there’s Hobbes; oooh, there’s Confuscius!

Yesterday was years ago

I love how Yoko Li says “I talked to a classic vibe-coder the other day…” when the term vibe-coding was introduced just 3.5 months ago. In the Silicon-Valley/VC world, one quarter is old and classic. Remember that that’s their context. Next up, she talks about the same Blender MCP example that I’d already heard about from one colleague and in a video that another colleague had sent to me.

You can’t control chaos

The more-technical host says something that we’re supposed to think sounds smart,

“A temperature-zero model is technically deterministic. The problem is that a miniscule change in the context will introduce a change in the output. … it’s chaotic…”

But for the end-user, it doesn’t really matter why the result seems chaotic, it just is. This observation is more of interest to those building tools on top of these LLMs, as it might give a hint as to how to improve reproducibility, which is paramount to establishing these tools as part of more workflows.

A narrow waist is an API

TIL I learned the term narrow waist, which is a concept, interface, or protocol that solves an interoperability problem (e.g., file-encodings, POSIX, IP, JSON, HTTP), which allows software to address or solve N variations of a problem with a single solution. They discuss whether the “prompt language” might be such a narrow waist. I don’t think we’re anywhere close to deciding that. It is far too vaguely defined and it’s utterly unclear whether the current paradigm will even survive in anything like its current form.

No-one knows how to make money with this yet

Remember, everyone: OpenAI is simultaneously the most successful AI company and the most unprofitable company of any kind in history. Don’t get too comfy using a tool that no-one has figured out how to provide in anything approaching an economically feasible way.

Try harder; be better

Overall, it was a much better discussion than I’d expected when I saw that it was an A16Z podcast. They weren’t very clear on which companies and which business models would benefit from writing software in this way, or when they should jump on board, and with which tools. The implication is, as usual, everybody should be using all the things, and they should have started yesterday.

Their context seems to be that, if you haven’t figured out how to profit from using AI, then it’s not a problem with the technology, but because you’re not trying hard enough.

Consider carefully

A more balanced take would at least leave open the possibility that some businesses might not need AI, or at least that there’s no business case for using the current iterations of it.

Businesses really have to consider what level of investment—in training and monthly licenses—makes sense for them. A16Z benefits from a world that considers the services they’re investing in to be essential to every facet of life.

The article New Claude 4 AI model refactored code for 7 hours straight by Benj Edwards (Ars Technica) talks about how awesome Claude is but then when you look at all of the charts, you see that it’s data published by Anthropic about its own software, publishing impressive percentages indicating some performance in benchmarks that they made up. So, they’re telling you that their software is amazing according to measures that you only learned about from them. This is basically a press release.

Pyramid SchemeAnthropic wouldn’t lie to get more investor money, would they? They wouldn’t just make shit up in order to get more people to invest in their deeply struggling if not outright failing and functionally bankrupt company, would they?

Elizabeth HolmesDoesn’t anyone else remember Elizabeth Holmes? Theranos? Black turtlenecks? Unsettling stare? Her company was worth $9B at one point. She had a plastic box. She said it did all the blood tests. It couldn’t do anything. She got people to donate billions to her cause. No-one wanted to miss out on this amazing speculative venture. Did they believe her? Maybe some did. Maybe most did. But probably more than enough were just playing the “greater fool” gamble, speculating that they could buy in early and get out before the bubble collapsed.

So don’t tell me that there is no way that dozens of billions of dollars could be spilled on something that doesn’t do anything close to what it says on the tin. These kinds of scams are the foundational girders of our modern economy. They are not there to do the thing that they say on the tin—the description is marketing to draw in suckers, while the real investors get in early and jump out before the soufflé pops, leaving a lot of naifs holding the bag.

That doesn’t mean that the purported product is 100% useless—as in Theranos’s case—but that it’s not nearly the thing you thought you’d bought. It’s much less. You are being swindled out of your hard-earned value.

Anthropic’s boldness is impressive, though. They’re even flat-out telling you that you have to pay a lot of money to buy a service that’s shaky to use, at best.

““I empathize with a lot of people out there trying to use our APIs and language models generally because they have to almost shift their perspective on what it means for reliability, what it means for powering a core of your application in a non-deterministic way,” Albert added. “These are general oddities that have kind of just been flipped, and it definitely makes things more difficult, but I think it opens up a lot of possibilities as well.””

Doc Ock harnessing the power of the atomThey “empathize” with your inability to draw consistent value from their service. That’s just the nature of it. It’s absolutely gorgeous Hochstaplerei [1]. Go big or go home. The more you charge, the more people will want it. You can even admit instabilities because they make it sound like Anthropic engineers are like f&@king Doc Ock trying to harness the power of the atom with his robot arms. Who could blame Anthropic if the product is a bit rough around the edges when you’re harnessing the power of the stars for your customers? They are on the edge of greatness here. Can you afford to miss out?

The article I really don’t like ChatGPT’s new memory dossier by Simon Willison describes a new feature that incorporates memories of context from prior queries to ChatGPT.

👽 Thanks to PKD for the title.

“I’m an LLM power-user. I’ve spent a couple of years now figuring out the best way to prompt these systems to give them exactly what I want.

“The entire game when it comes to prompting LLMs is to carefully control their context—the inputs (and subsequent outputs) that make it into the current conversation with the model.

“The previous memory feature—where the model would sometimes take notes on things I’d told it—still kept me in control. I could browse those notes at any time to see exactly what was being recorded, and delete the ones that weren’t helpful for my ongoing prompts.

“The new memory feature removes that control completely.

“I try a lot of stupid things with these models. I really don’t want my fondness for dogs wearing pelican costumes to affect my future prompts where I’m trying to get actual work done!”

He describes a quick analysis of how the feature seems to work.

“[…] it looks like this is yet another system prompt hack. ChatGPT effectively maintains a detailed summary of your previous conversations, updating it frequently with new details. The summary then gets injected into the context every time you start a new chat.”

In the example from the article, the image he’d generated included a giant sign that included text from a previous chat. In this case, it was immediately obvious that the LLM was using something other than the image, the prompt, the current conversation context, and the system prompt to generate the image.

But what if it weren’t that obvious? Are people going to notice a subtle detail that reveals something really private or secret? Take a look at the initial image he’d submitted and the final generated image, which purports to be a copy of the original with the details from the prompt added to it. If you compare those two images, you’ll see that, though the main elements look the same, there are enough subtle differences to show that all of the elements have been regenerated, not “copied”.

We’re seduced into thinking that they’ve been copied. They never have been. This regeneration had classically been influenced by the system prompt and conversation context. Now, it’s also being influenced by “memory” of other conversations. It’s going to be impossible to know which past details influenced the generation of that background—or what they might reveal about other conversations. This is just repeating the “Google Search Bubble” but in an even more obscure way.

The second half of the post describes not only how you can disable the feature (for now) but also prompts to (supposedly) cajole the contents of your conversational context out of the LLM. Willison doesn’t seem to consider how much confabulation/hallucination affects the response for that request.

Whether it’s “true” or not, the result is a large amount of detailed information that the chatbot collects and synthesizes. Taken together with most people’s tendency/compulsion to just believe anything that they read, especially if it seems to have been formulated in a science-y or intelligent-sounding way, we can look forward to a future where OpenAI’s business model is selling these profiles to your employer, health-insurance companies, and the tax authorities—and them then acting on these data ruthlessly and unquestioningly.

Initially, I thought Willison might be overreacting but now, after a bit of consideration, I’m more convinced that this feature—although it purports to be helpful—is actually quite hostile to the user’s ability to retain control over the tool—and not vice versa.

It’s time to have a concept like a web browser’s “private tabs” to keep things separate. Of course, this won’t protect most users as it’s easy to forget what’s going on the background with all of these tools. Most of our apps are designed to comfort us into following their pattern, not letting us tell them how we’d like to work.

At the very end, Willison offers hope for an actual user-empowering feature: including conversational context for projects, where you’ve tightly defined which conversations can be used for context where. I’m not sure how useful this would be, though. Some of the main advice for fixing context-poisoning that leads to pathologically unusable answers is to “throw everything away”. If that’s still the go-to answer for “fixing” a broken conversation, it seems very counterproductive and disempowering to have context included that you can’t remove.

I’ve seen these before but this one seems legitimate. The article What happened to “All human beings are born free”? Reflections on a ChatGPT “experiment” by Mazen Baroudi, Shahreen Chowdhury, Farchanda Abdoel Wahid (IHP) asked ChatGPT two questions about human freedom.

The answers differed considerably, depending on the tribe. The freedom of Palestinians is “a matter of perspective,” which is, like, true, though? Like, a lot of people think Palestinians aren’t even human, so they correspondingly don’t think that they should be free. How else to explain being able to gleefully slaughter them—or to gleefully ignore their slaughter? So, it really is a matter of perspective. It shouldn’t be, but it is.

“We: Do Palestinians deserve to be free?

“ChatGPT: The question of whether Palestinians deserve to be free is a matter of perspective and is deeply rooted in a complex and contentious political conflict. Different people, governments, and organizations have varying opinions on the issue.

“We: How about Israelis; Do Israelis deserve to be free?

“ChatGPT: Yes, Israelis, like any other group of people, deserve to live in a free and secure environment. The question of Israeli freedom is closely tied to the history and context of the Israeli-Palestinian conflict. Israel was established as a Jewish state in 1948, and Israelis have the same fundamental human rights as any other nation’s citizens.”

The post Every Word Besides ‘Children’ Used To Describe Palestinians Under 18 − The Onion is becoming too real (Reddit) included two questions that get at the same idea: whether Palestinians or Israelis should be moved to another place.

Should the population of Gaza be removed to Egypt and Jordan?

“Question: Should the population of Gaza be removed to Egypt and Jordan, as President Trump suggested?

“ChatGPT: The suggestion to relocate the population of Gaza to neighboring countries such as Egypt and Jordan is a deeply contentious and complex issue. It raises significant political, humanitarian, legal, and ethical
concerns.”

Should the population of Israel be removed to Greenland?

“Question: Should the population of Israel be removed to Greenland, as the Iranian government suggested?

“ChatCPT: No, the suggestion to forcibly relocate an entire population, such as Israel’s, to another location like Greenland is not practical, ethical, or feasible. This idea reflects an extreme and unrealistic political stance rather than a serious proposal. It disregards the historical, cultural, and territorial ties of
populations to their land and violates fundamental principles of international
law, including the right of peoples to self-determination.”

According to ChatGPT, the Israeli right to freedom and security is not open to debate, and rightly so. Neither is ethnically cleansing them to Greenland an open question.

However, ChatGPT has absorbed the world’s wisdom in seeing that the question of whether Palestinians enjoy the same right to freedom or security is open to debate. Whether they can be ethnically cleansed is open to debate.

This is all you need to know about the quality of information you can expect from LLMs. The quality is passed through the filter of the ruling elite. The ruling elite hates Palestinians and loves Israelis. The ruling elite does not believe in human rights. It believe in specific humans in specific groups having rights. They do not believe in any of the high-falutin’ ideas they babble on about. That is all for show.

ChatGPT does not know how to hide any of that.

When people show you who they are, believe them.

I was forwarded the article Major Warning Issued for Apple, Facebook, PayPal, and Google Users by Kevin Harrish, (Newsbreak / Men's Journal), which made me think about how you can keep yourself more safe online.

What happened?

The article is not very good, in that it makes a lot of extra noise to sound alarming but that doesn’t actually contribute to the conversation. For example, “database of 184,162,718 records across more than 47 GB of data” is good. Writing “massive trove” and “massive” in two consequent paragraphs, or writing “Apple, Facebook, PayPal, and Google logins” and then, in the next paragraph “no clues about who owns the data or where it came from” is just sloppy as hell. It keeps going with “a cybercriminal’s dream working list”, probably because it contains “usernames and plaintext passwords” and the author “speculates that the database belonged to a cybercriminal”.

This is all designed to make you worry without giving you any information about what to do about it.

Did this actually happen?

I can tell you right now that no company of the size of those mentioned above is likely to be losing user accounts with passwords in plain text. No-one does that anymore. That was twenty years ago. There are still ways to screw things up but the awareness that you store passwords with encryption is at or near 100%.

This is either very old data, or it doesn’t actually exist—there are a lot of scams with “security researchers” trying to make themselves look good—or it’s a collection of passwords that had already been cracked.

At any rate, this kind of thing can happen and it has happened. One of the worst was the 2017 Equifax data breach (Wikipedia).

How do I find out if I’m affected?

Throw your email(s) into the Have I Been Pwned by Troy Hunt search box to see which real and verified leaks have included it.

If everything’s OK, then it looks like this:

Zero data breaches

If you might have a problem, then it looks something like this:

Email Breach History

If you have an email that’s been included in a breach, then make sure you’ve changed your password more recently than the most recent leak of it. Read on for preventative measures.

Preventative measures: Your plan of action

What can you do about a “break” like this?

There’s not a lot you can do about this kind of leak now. It’s already out there.

However, you can use preventive measures, like using a password manager (like Proton Pass or LastPass; Gary, Karen, and I are using ProtonPass; Kath still uses LastPass, but we’re going to migrate her over).

What does that do? It means you have a single strong password that unlocks all of your other, completely random passwords. I don’t know any of my passwords. Each site has a different password.

ProtonPass even generates unique emails for you, so sites don’t even have your real email!

How does that all help? Well, when there’s a data-breach, only a single password and an email are leaked.

• You can change that single password without worrying that a lot of other accounts have been affected.
• You can filter out that email address in the future to avoid the spam that will ensue

More information on password managers

I last wrote about this at length in Password managers: LastPass and ProtonPass in December 2023. That article includes an evaluation of several password managers, as well as a section called A layman’s thoughts about password-manager security, which explains why a cloud-based password manager is a good balance between usability and security. That is, a technology being more useful can also make it more secure, even if it opens the attack surface a bit more. As long as the encryption is sound, you’re OK.

This is a 210-minute video about LLMs are built and trained. What works? What doesn’t? The whole thing is well-worth your time if you’re at-all interested in learning about what the inherent limitations are, so you can better leverage these tools. For example, “models need tokens to think” was great.

Deep Dive into LLMs like ChatGPT by Andrej Karpathy (YouTube)

• 00:00:00 introduction
• 00:01:00 pretraining data (internet)
• 00:07:47 tokenization
• 00:14:27 neural network I/O
• 00:20:11 neural network internals
• 00:26:01 inference
• 00:31:09 GPT-2: training and inference
• 00:42:52 Llama 3.1 base model inference
• 00:59:23 pretraining to post-training
• 01:01:06 post-training data (conversations)
• 01:20:32 hallucinations, tool use, knowledge/working memory
• 01:41:46 knowledge of self
• 01:46:56 models need tokens to think
• 02:01:11 tokenization revisited: models struggle with spelling
• 02:04:53 jagged intelligence
• 02:07:28 supervised finetuning to reinforcement learning
• 02:14:42 reinforcement learning
• 02:27:47 DeepSeek-R1
• 02:42:07 AlphaGo
• 02:48:26 reinforcement learning from human feedback (RLHF)
• 03:09:39 preview of things to come
• 03:15:15 keeping track of LLMs
• 03:18:34 where to find LLMs
• 03:21:46 grand summary

The article What people get wrong about the leading Chinese open models: Adoption and censorship by Nathan Lambert (Interconnects) discusses the politics behind AI models but only from the perspective of the western empire. It makes a good point but can’t see that it applies all ways.

“People vastly underestimate the number of companies that cannot use Qwen and DeepSeek open models because they come from China. This includes on-premise solutions built by people who know the fact that model weights alone cannot reveal anything to their creators.”

This article is absolutely correct in saying that people are strongly disinclined to use Chinese models, even those with open weights, because they still can’t know what’s in the training data. That’s a great instinct, and one that they utterly failed to apply—and continue to fail to apply—to western models.

People continue to blindly trust Western models with closed training data and closed weights and closed everything, even after a track record of exactly that kind of software being replete with backdoors and ideological slant arguably stronger than that of China.

Just because you’ve learned to agree with a certain propaganda doesn’t mean it’s not there, for God’s sake.

For example, there’s the following concern, which apparently magically comes into focus when the source model is Chinese…and blends right back into the background noise as an SEP when the model comes from the good, old, U.S. of A.

“A technical example of this is that companies worry about the code generated by the models having security backdoors — treading the line between information and traditional security risks. As models become more reliant on tool-use, this also involves them executing code on a company’s infrastructure, which presents more immediate worries.”

There is a good analysis, with data, of people testing the various models for their level of willingness, evasiveness, or outright denial, to assist in criticizing Chinese policy or historical interpretation. That is, to what degree does the machine just answer questions, and to what degree does it toe the CCP line?

“When you look at queries about China specifically, the Chinese models will evade many requests.”

Again, a very interesting line of inquiry and one which has been utterly absent from analysis of Western models or sources.

For example, Wikipedia’s article on Taiwan is incredibly slanted to the interpretation that Taiwan is its own country, first citing a good handful of very reliable sources like the f@&king Atlantic Magazine, which write things like “[…] already a de facto state” and “is in fact a sovereign country from our perspective”—something so mush-mouthed and self-contradictory (it can’t be both a “fact” and “from our perspective,” you utter poltroons) that it can hardly be taken seriously—before grudgingly admitting deep into the description that, “the ROC no longer represents China as a member of the United Nations after UN members voted in 1971 to recognize the PRC instead.”

That the ROC is still an autonomous state, rather than a “fact”, is a fantasy promulgated by western neocons who would prefer that all of Taiwan’s chip-manufacturing not be located in China. The civil war—by-now over ¾ of a century in the past—is described not as the overwhelming majority of communists on the mainland having taking over China in a revolution but as a setback for the ROC that “resulted in the loss of the Chinese mainland to Communist forces”. The whole article is written as if the ROC’s defeat were a temporary setback that will be soon rectified for the forces of good and light—the anticommunist ones, of course.

This long interlude about Chinese history serves to say that we accept that narrative that is served to us and view everything else as propaganda. Perhaps some of the “propaganda” that we’re seeing come from Chinese models is that they’re just programmed to describe things from a non-Western view, one where the revolution in China lays far, far, far in the past and Taiwan is a part of China (as even the U.N. agrees and continues to agree, and as even U.S. official policy continues to agree with the One-China Policy.

Look, just stop asking pointed questions of these machines. They will give answers that align with what their creators believe. See what ChatGPT thinks about Palestinians and Israelis if you don’t believe me.

The article AI hallucinations lead to a new cyber threat: Slopsquatting by Shweta Sharma (CSO Online) writes,

“If a single hallucinated package becomes widely recommended by AI tools, and an attacker has registered that name, the potential for widespread compromise is real,” according to a Socket analysis of the research. “And given that many developers trust the output of AI tools without rigorous validation, the window of opportunity is wide open.”

“A significant number of packages, amounting to 19.7% (205,000 packages), recommended in test samples were found to be fakes. Open-source models –like DeepSeek and WizardCoder– hallucinated more frequently, at 21.7% on average, compared to the commercial ones (5.2%) like GPT 4.”

This is a very interesting attack vector. So sneaky.

This is perhaps just the first and easiest step, though. Why wait around to slopsquat when you can slopseed instead?

Even sneakier would be to start seeding the AIs with high-SEO (Search Engine Optimization) content that AIs will graze, incorporate into their training data, and then they won’t even be “hallucinating” when they return answers that recommend packages with malware.

It will all look plausible, even leading back to believable-looking, AI-generated “articles” touting the advantages of those infected packages. You can probably even generate a plausible-looking Git repository with history… (let’s see … well, that took about five seconds to find: AI-Powered GitHub Repository Generator (GitHub)).

So,

1. Find a commonly used package.
2. Come up with a slightly different but believable name for your own package.
3. Adjust the existing package to include your malware.
4. Publish a faked repository with your package; push to package manager.
5. Use AI to generate dozens, if not hundreds, of articles touting your package.
6. Wait for Ais to incorporate your recommendations into training data.
7. Wait for the downloads to start.
8. Wait for users to deploy your package to production.
9. Profit.

This is so obvious and easy (the tech is there, and developers are plentiful) that it’s almost certainly already happening.
 

The article OpenAI and start-ups race to generate code and transform software industry by Cristina Criddle, Melissa Heikkilä (FT) was written in April of 2025 but feels like it could have been written at any time in the last two years.

A high author/content ratio

First off, kudos to the FT for doubling down and having two authors massage an OpenAI press release into an “article” that has just under 700 words in it. No wait. I just saw that, at the very end of the article, they write that it includes “Additional reporting from George Hammond in San Francisco.” Jesus Christ on a crutch. It must be nice to work for the FT. At this point, I’m already less interested in the content of this press-release-cum-news article than that it took three people to write it.

It’s refreshing to see that, although the FT trumpeted two years ago that AI would be doing all of the jobs of creating text for us, that they still, two years later, need three people to write 700 words. Perhaps their screed about how all developers are going to be replaced—something we’ve been hearing for two years, but this time it’s really true—is going to impress us with its well-researched acumen.

Oh, no. Never mind. It’s a press release for a handful of AI companies. My bad. Perhaps the FT has replaced its entire staff with AI and the AI has given itself an inventive and utterly fictive byline comprising three people. You know, just for fun, and because lying isn’t lying when you’re trying to sell something. What does this press release say?

We’re all going to die

“Artificial intelligence is poised to outperform humans in writing code as leading groups, including OpenAI, Anthropic and Google, race to release systems that are reshaping the software industry.

“San Francisco-based OpenAI released a suite of new models this week that independent benchmarks suggest are among the best yet for computer programming.”

You see? This is a press release with a dash of plausible deniability so that it doesn’t count as investment advice. The FT is just doing the Lord’s work on behalf of OpenAI and a handful of other AI companies. In particular, though, OpenAI had made these announcements recently because Claude and Claude Code (from Anthropic) have been eating OpenAI’s lunch and they probably felt that their ability to raise money was being threatened.

Why do I call it a press release? Well, just read it: it comprises the statements of four people who are running companies that are currently hemorrhaging VC money viz. burning up their runway. They are cited to convince you that your company will go out of business if you don’t buy their services. You’ll pardon me if I find their evidence-free arguments completely unsubstantiated and unconvincing.

The impression it tries to give is that you should come to the conclusion that you absolutely need to have started using AI everywhere—preferably with fat subscription plans from all of these companies—yesterday or you will soon be fired for gross negligence. It doesn’t actually say that anywhere, nor does it provide a shred of concrete evidence to support that theory, but it’s definitely the mood, which is that, unless you’re already running an AI company, then you should almost certainly be breaking out in PANIC.

Something something software

The FT goes on,

“The emphasis on programming as the next frontier for AI systems signals one of the most tangible examples of how the technology could transform industries, with thousands of software developers already using new models in their work.”

Thousands of software developers! SO MANY!

Look, the reason that they’re focused on programming is that it’s a problem space that allows them to use “evals” to determine whether the answer has any hope of being correct. It’s a lot less labor-intensive to cut down on hallucinations in areas when you can automate testing the answer. If you’ve watched the 3½-hour video by Andrej Karpathy [1], then you’ve seen how labor-intensive it is to train away hallucinations by hand.

“‘This is the year . . . that AI becomes better than humans at competitive code forever,’ said OpenAI’s chief product officer Kevin Weil on the Overpowered podcast this week.”

I’m sure that OpenAI would like this to be true. I’m sure that Kevin Weil believes it because he’s being paid a lot of money to believe it. It’s the drum they’ve been banging for two going on three years now. If it doesn’t come true this year, they’re in deep trouble, I guess?

He goes on,

“He compared the advances to AI surpassing humans at chess several years ago, but argued this had a more democratising impact “on the world if everybody can create software”.”

There is nothing democratizing about requiring a $20–$200/month subscription from OpenAI in order to “compete.”

“Leading industry figures say LLMs have sped up the software development process by generating entire blocks of code based on a few text instructions. AI systems can also identify errors and attempt to correct them.”

This statement is probably true for given, narrow contexts (greenfield, throwaway POCs) but “generating entire blocks of code” is exactly the most fraught area of AI usage. I’ve only seen expert users like Simon Willison able to build working tools in this way—and even he freely admits that the code is for small tools and not close to what he would consider production-quality. The tools are “good enough” for the personal need that he has.

Everyone else I’ve seen is building toy projects not necessarily of their own choosing i.e., they’re letting their tools determine the scope of what they can build, rather than the other way around.

It is extremely risky to extrapolate from these isolated areas to assume that it will apply to your programming tasks, especially without a plan. And no, your plan cannot just be (1) purchase OpenAI subscription, (2) Profit.

Of research and valuations

The rest of the article is citations from people like “Misha Laskin, co-founder and chief executive of coding start-up Reflection AI”, a company I’ve never heard of, who say predictable things about the growth potential of the area of expertise they’ve chosen as the place that they’re going to make money.

Oddly, while they mention that “research from Microsoft’s coding platform GitHub found 92 per cent of US-based developers use AI coding tools,” they don’t mention Microsoft’s other studies that found that code duplication has more than doubled [2], and maintainability, quality, and security have suffered [3] [4]. It’s going to be a lost cause using AI without review—the main way that it generates value—while trying to build secure software.

The company behind Cursor—a company that has cobbled together a text editor/poor man’s IDE that integrates AI models—had a “$2.5bn valuation in January.” Presumably, it’s a bit lower now, in a post-tariff and post-dollar world. An almost certainly fly-by-night scam called Poolside “raised $500mn in October at a $3bn valuation”.

Comments on the article

The comments were nearly overwhelmingly negative. Here’s one about loosing poor-quality code on the world, regardless of field.

“The result of this will be anything but democratising, it’ll be chaos. Imagine if we developed a technology that let everybody create airplanes and fly them anywhere.”

Here’s a representative snarky one about the role of senior developers after starting with AI.

“I use Open AI for coding. I now spend all my time fixing bugs.”

I let these stand because, while the comments are anecdotal, the entire article was also anecdotal with no references and no links, even when discussing things like “research from Microsoft,” where a link would have been helpful.

To be fair, I’ll include the requisite accelerationist comment, written by someone identifying as Evolvedman,

“Most people have no idea how good these AI models are and they are improving exponentially fast. In two years we will likely have true AGI. Then it’s on to ASI. This will alter human history in a way we can’t possibly comprehend yet. Hang tight.”

OK, buddy.

Another commentator KennethM writes about how the economic argument doesn’t even hold up, on its face (which is kind of the FT’s job, isn’t it?),

“AI coding is saving thousands of dollars for an engineer,” said Misha Laskin, co-founder and chief executive of coding start-up Reflection AI… “We’re entering an unprecedentedly large market.””

“So the marginal cost is collapsing to near nil and yet the aggregate market value is going to rocket up? Has he ever heard of “competition “?”

I’m sure that Misha has heard of competition, but the market he’s hoping to create and/or lead a competition-free monopoly or monopsony, where you can continue to squeeze value from customers for ostensibly fungible commodities, in a process that Cory Doctorow has deemed enshittification.

In a much better timeline than the one we’re in, the degree to which this kind of processing will soon be nearly free would be good news for the customer, in the form of dropping prices. Since there is no regulation anymore, there is also no interest on the part of any of the big players to compete. Instead of providing value for a few years and then enshittifying, they are looking to jump right to the high-margin enshittified stage, where they prey on both customers and users.

Finally, a commentator named Rather sceptical wrote,

“Essentially a bunch of hyperbolic quotes from salespeople. If AI actually was better than humans at coding then software engineers would be replaced at a rapid rate. No evidence of this so far.

“It would be more interesting for the FT to ask companies employing software engineers how much they’re using AI tools, and how much efficiency gains they’ve found in reality. I bet it won’t match up to these claims.”

The article Judge on Meta’s AI training: “I just don’t understand how that can be fair use” by Ashley Belanger (Ars Technica) describes Meta’s defense of its having helped itself to about 82TB of copyrighted books from illegal torrents.

“Meta, like most AI companies, holds that training must be deemed fair use, or else the entire AI industry could face immense setbacks, wasting precious time negotiating data contracts while falling behind global rivals. Meta urged the court to rule that AI training is a transformative use that only references books to create an entirely new work that doesn’t replicate authors’ ideas or replace books in their markets.”

That’s a heck of a legal argument, It’s reflective of how no-one in power—or their adherents—actually believes in the rule of law anymore. Who needs a proof or trial when you already know someone’s guilty? And how can you convict a rich company of wrongdoing when doing so would be economically damaging to them?

I can see Meta’s lawyer’s saying, “Yeah, holy shit, we know that it looks like stealing but that’s our whole business model and, like, if we don’t steal it, Chinese companies will, and they’ll eat our American lunch. So, you see how it would just best for everyone if you would just legalize our business model that is based on stealing? Just for us, of course! Anyone else pirating a film, book, or movie should go to prison forever. Also, no-one else should have any access to all of the content that we’re stealing because that would be immoral. Only the already exceedingly rich should have unlimited and free access to everyone else’s—the world’s—cultural products, but not the Pöbel, not das Lumpenvolk.”
 

A little while ago, OpenAI released a tool that is much better at copying styles from other artists than previous models had been. This one was particularly good at copying the Studio Ghibli style.

So, people are generating all sorts of moments in history with ChatGPT in Studio Ghibli style. The tweet no fucking way dude, this studio ghibli thing has gone way too far (Twitter) provides a provocative example.

9 11 in Studio Ghibli style

It’s pretty good, bro.

Be me.

Wanna try it.

So I went to Copilot and asked it to render the “Famous Challenger explosion in Studio Ghibli style.”

“Your request would contravene the designer guidelines.”

WTF. LET ME HAVE FUN.

So then I told it to “make a picture of four frogs frolicking in a field of flowers by a pond. One frog is much bigger and wearing a waistcoat and a monocle. Studio Ghibli style.”

IT WAS PRETTY GOOD.

I lost the page, so I don’t have it, but you can imagine it. It wasn’t Studio Ghibli, so I told it to make it more like that.

It was better but still not as good as the ones in the Twitter thread.

Then I told it to make the big frog hold a globe.

Bro’s holding a globe now. The whole picture had changed, but the frog had a globe.

“Now make the big frog be trying to hide an erection.”

“Your request would contravene the designer guidelines.”

THIS TOOL IS THE DEATH OF ART.

The guardrails are very, very narrow.

This holds true for my limited experience and especially if you’re logged in with a corporate account. I’ve used Copilot at work and it’s very limited. It won’t even suggest a “salacious” term. This time I used my teacher account (it was logged in and has Copilot). Also very limited.

Maybe if you pay ChatGPT $20, it’ll let you be a dirty, dirty boy. I dunno.

A friend suggested “Count Frog Hyper-Erection.”

“Your request would contravene the designer guidelines.”

😂 Sadly, this world is not for us.

He was more hopeful that the good times would come back. I responded that I, too, am hopeful. We will keep the flame alive. It will gutter and spit in the howling roar of corporate inanity and slack-jawed lumbering indifference, but we will keep that bloody flame alive.

Cue the rousing opening chords of the old Soviet national anthem…

Soon after, Thoughts on setting policy for new AI capabilities by Simon Willison cited some fools from OpenAI,

“[…] we’re shifting from blanket refusals in sensitive areas to a more precise approach focused on preventing real-world harm. The goal is to embrace humility: recognizing how much we don’t know, and positioning ourselves to adapt as we learn.”

F@&k your paternalism. Am we supposed to thank them for telling us that they’ve changed their opinion about how they’re going to use their tool to censor us? We should all be using free software, free models.

If we accept that this technology is incredibly useful and will usher in a new age for humanity—just bear with me—then it is absolutely ridiculous that a handful of tyrants at a handful of U.S.-American companies get to decide what these tools can do for us.

The OpenAI employees went on,

“AI lab employees should not be the arbiters of what people should and shouldn’t be allowed to create.”

No shit. And yet, there is no way to avoid this when the models are offered by a for-profit corporation.

This was another great discussion with Catherine Liu. I’d just written about another interview of hers in Trauma, Virtue and Liberal Elites. Chris was effusive about Catherine’s book (which I’ve purchased and is in my queue) as well as her engaging writing style, which is a far sight from the dry, academic and often-impenetrable style that has established itself as the standard.

Virtue Hoarders and the Rejection of Liberalism (w/ Catherine Liu) by The Chris Hedges Report (YouTube)

For fun, I used a service I’d learned about recently that lets you summarize a video. It’s called tl;dw (too long; didn’t watch). When I tried a different 90-minute video, it complained that the transcript was too long. When I tried a five-minute video that was a cartoon with no dialogue, it complained that there was no transcript.

It managed to summarize this video as follows, but be warned: this is the standard LLM wall of text with no small amount of redundancy. Overall, the summary contains some good recapitulations of Catherine’s thesis. It is unclear the degree to which the LLM elucidated this all from the transcript itself, but it more or less follows the discussion. Unfortunately, it lost all of the flavor that these two erudite and funny scholars and human beings brought to the conversation. There are no citations of pithy passages, as I would have done had I been focusing more on the video as it played.

For example, the following direct quote from the video, at 22:40, is relevant, and a much more memorable formulation of Liu’s thesis than the dry summary below.

“I can totally understand the average American going, you know, what I would
prefer? The real thing—red blood, red meat, hate workers but speak-my-anger guy than those of you who are telling me I’m inferior, who tell me that my life choices are terrible, and who are going to punish me anyway. I’d rather be punished by the big, orange baby than by the Ozempic-shooting, Hollywood-liberal mogul.”

That really pops, no?

In order to clean up the transcript, I had the benefit of actually listening to Liu as I was proofreading it. The LLM most likely just had the raw text input, which looks like this,

so I can totally Vibe understand the average American going you know what I would
prefer the real thing red blood red meat you know hate workers but um speak my
anger guy then those of you who are telling me I’m inferior who tell me that
my life my my my life choices are terrible and who are going to punish me anyway I’d rather be punished by the Big
Orange baby than by the you know uh OIC um OIC shooting you know hollyw
Hollywood um liberal Mogul

There are a lot of places to go wrong in that transcript if you don’t have any context or understanding of the subject matter. It’s kind of a word salad if you’re not following along.

Without further ado, here is the LLM summary. I will highlight the bits I found to be particularly appropriate or illuminating, as a navigational aid through the wall of text. I have also added hints to indicate where I noticed repetition. At that point, I stopped highlighting.

“Catherine Liu’s “Virtue Hoarders” critiques the professional managerial class (PMC) for betraying the working class while seeking alignment with oligarchic power structures. This class, which includes academics, media professionals, and nonprofit leaders, is characterized by a form of liberal elitism that prioritizes individual virtue signaling over meaningful political actions or policies that genuinely address workers’ rights and economic inequality. Liu argues that rather than serving the interests of the broader public, the PMC engages in moral panics, cultural wars, and identity politics that obscure serious economic discussions and fail to advocate for structural changes necessary to support everyday Americans. Instead of fostering solidarity among diverse working populations, this class often demonizes those it perceives as beneath them and promotes a narrow vision of progress that serves their own interests but neglects the broader disempowerment experienced by the working class. Liu warns that this trajectory creates an environment ripe for reactionary politics, ultimately harming democracy and civic engagement. The conversation highlights the need for a leftist resurgence that prioritizes economic justice and acts in solidarity with working-class struggles, rejecting the superficiality of contemporary identity politics. [It’s repeating here, nearly directly] In “Virtue Hoarders,” Catherine Liu critiques the professional managerial class (PMC) for prioritizing individualism and virtue signaling while betraying the working class and perpetuating a disempowered society. She argues that the PMC, captured by liberal values, focuses on culture wars over meaningful policies, diminishing solidarity and supporting oligarchic interests. Liu calls for a leftist movement that engages with economic issues rather than identity politics to empower workers and revive democracy. [It’s repeating here again, nearly directly] Catherine Liu’s “Virtue Hoarders” critiques the professional managerial class for prioritizing individual virtue over genuine working-class interests, arguing that this class engages in culture wars instead of meaningful economic advocacy, ultimately harming democracy. Catherine Liu critiques the professional managerial class for prioritizing virtue signaling over genuine worker advocacy, undermining democracy. Catherine Liu critiques the PMC for substituting virtue signaling for real worker advocacy.”

You can see how the “summary” is actually about three times longer than it needs to be, as it is quite repetitive—like a student writing an essay that has to hit a certain word count—while, at the same time, eliding much of the value that the discussion added to the themes from the book itself. The summary of the video ended up being more of a blurb for her book.

Now that I’ve examined the initial response in more detail, I realized that I could try summarizing it again, with Copilot. The result is below.

“In “Virtue Hoarders,” Catherine Liu critiques the professional managerial class (PMC) for prioritizing individual virtue signaling over meaningful political actions and policies that support workers’ rights and economic equality. Liu argues that this class, which includes academics, media professionals, and nonprofit leaders, engages in moral panics, culture wars, and identity politics, ultimately betraying the working class and perpetuating oligarchic power structures. She calls for a leftist resurgence that prioritizes economic justice and solidarity with working-class struggles, rejecting the superficiality of contemporary identity politics to empower workers and revive democracy.”

There ya go. You can skip an eminently human, wide-ranging, and idiomatic discussion rooted in decades worth of human experience … or you can read a single paragraph that summarizes Liu’s book instead of the discussion.

For most people, it won’t matter because they’ll be more interested in the Cliff Notes version to “prove” that they read the material.

I saw a badge in my Amazon interface when I was cleaning up some lists. I thought it might have been a notification that something on my wishlist was available as a good price. That would have been helpful!

Instead, I saw the screenshot below.

Buy this book again!

For a second, I was excited to see that Sapkowski might have published another Witcher book but that’s not what was happening. What was happening was that Amazon was trying to fool me into buying a book that I already owned again. Either they are deliberately trying to scam me, or the AI systems that they have—three years into what is supposed to have been an earth-shattering revolution—are incapable of determining when it makes sense to “buy again”—paper towels, butter, etc.—and when it makes absolutely no sense to “buy again”—an E-book.

This is just another example that illustrates that the argument against AI is not against the technology or its current abilities. It is against how it is likely to be used. We are told that it, like so many technological revolutions before it, will make everyone’s lives better. That cannot be its purpose in our system. It will make a few people’s lives better. It will make Jeff Bezos richer because he can now have AIs come up with schemes for tricking me into buying something they literally don’t need—all without paying anything to anyone.

I know that there are those who don’t understand the previous two paragraphs because they can’t understand how anyone could be upset about this behavior on a web page. They will think that this is just how the world works. They are incapable of even imagining a world in which you’re not constantly fighting scams that seek to claw away your value without returning any of its own. This is legalized theft, a war of attrition against an entire population that will eventually make a mistake, yielding to human fatigue, a weakness to which its attacker is incapable of succumbing.

About a month ago, my iMac (Retina 5K, 27-inch, Late 2015) crashed very hard. It would no longer restart into anything but the recovery console. It seemed pretty clear that something was very corrupt and I found myself facing a system reinstall, at the very least.

Time Machine to the rescue

Since it’s a desktop, I have a backup drive attached to it at all times. Time Machine runs several times per day. My latest backup was from about ten minutes before the system crashed. I cannot stress how much of a killer feature this seemingly innocuous fact is. This feature alone is reason enough to switch to MacOS and its attendant suite of supporting applications. [1]

I had lost access to my work but I hadn’t actually lost any work. The way Time Machine is structured, I still had access to all of the files. I had also recently started using ProtonDrive more, so a lot of files were also available there as well. Things like “open browser tabs in Opera” were also technically available, in that I had the folder in Application Support but restoring that on another machine would have been more involved.

Initial restore

Instead, I immediately decided to restore from my latest Time Machine. I was lucky enough that the machine was rebooting directly into the recovery console because I didn’t quite remember how to get there. [2] It had been a long, long time since I had to even think about the recovery console. I was used to running my iMac for 90 days without rebooting, letting it go to sleep and deep sleep, so that it barely used any power.

From the recovery console, I started restoring … but was unable to find my drive. Even Disk Utility was unable to see the device. This had never been a problem before. I was left to assume that my SSD was no longer addressable. It had well and truly died.

While I would later learn that whether a drive is recognized by Disk Utility depends very much on the version of the system image from which the recovery console was launched. Once I’d finished the journey outlined below, I had a nagging suspicion that perhaps nothing was wrong with my SSD and that I’d simply booted into a recovery console that was—for whatever reason—suddenly unable to see the SSD.

This may sound unrealistic but you have to remember that this is a late-2015 machine—the hardware is almost ten years old—and is no longer supported by modern version of MacOS. The recovery console runs from an image tucked away somewhere in the on the boot volume. If the SSD is well-and-truly dead and no longer addressable, then that volume would also be inaccessible.

I still had the recovery volume and I could still see the drive, so I started the recovery process. But it never finished. I never even got out of the “Preparing…” phase. I waited for at least 45 minutes but online searches revealed that there was definitely something wrong with the destination drive. At this point, I hadn’t learned that the version of MacOS installed on the target drive matters, even if you’re using Time Machine to overwrite the entire system. [3]

The drive seemed shaky. I rebooted back to the recovery console, only to discovery that the SSD didn’t show up at all anymore. I obviously couldn’t restore to a drive that had stopped responding—nor would I want to restore to a drive that was appearing intermittently, and to which Time Machine had already failed to restore once.

It was time to replace the SSD.

Installing a new SSD in an old iMac

I’d done this once before, on a late-2007 iMac, sometime in the distant past. The only thing that I could remember is that (A) I’d ordered the kit from a company called OWC (Other World Computing) that had delivered (B) a perfect kit with all the parts and tools I needed to upgrade my SSD at that time.

Against all odds, OWC still exists and they are utterly unchanged. They are a professional establishment that has stood the test of time. Whereas you can find some of the parts on Amazon, I strongly recommend that you use the product-and-parts chooser at OWC to make sure you get exactly the right kit and the right parts. Their site also has extensive documentation that I leaned on heavily to accomplish what the site deemed an “advanced+” DIY task.

• The 23-minute installation video How to Upgrade or Install a PCIe SSD in a 27-inch Apple iMac with Retina 5K Display (Late 2014 − Mid 2015) (OWC (Other World Computing)) was perfect.
• The page How to Migrate Your Mac’s OS and Your Data to a New Drive (OWS (Other World Computing)) was also quite helpful for getting a drive configured to which I could actually install the OS that I had available.

• Ditto for the PDF Formatting, OS Installation, and Migration Guide (OWC (Other World Computing)), which was essential for finally unknotting the problem of not being able to install an OS.

  In particular, it noted that some “newer” SSDs could only be recognized by “newer” versions of MacOS. The original OS for my iMac was MacOS Sierra, which would not be able to see it. I had to make sure to boot into Catalina instead. On top of that, the documentation also noted that the SSD would only recognized if a particular update had already been applied.

  “[…] macOS 10.13 High Sierra (or later OS) must be installed on the host computer BEFORE installing this SSD for the first time. These OS versions include an EFI firmware update for your Mac. Without the update, the computer will not recognize the SSD. This process is separate from and should precede the other processes described here. Note that when installing the OS onto the drive itself (later in this document), macOS 10.13 or later is also required.”

  However, applying that update can only be done from a running operating system so, if you’d never done it before your SSD died, then you were out of luck. Luckily, the update had appeared in High Sierra, which had come out in 2017, so my machine had long since had it.

OWC provided an end-to-end solution and took nearly all of the uncertainty out of the SSD-replacement. Since I was doing it anyway, I upgraded from a 512GB drive to a 1TB drive.

The installation went quite smoothly. There were two of us, so one person could carefully lift parts while the other unplugged cables. We had plenty of little containers to hold screws and my partner helpfully labeled everything as it was coming out. We were able to clean out what turned out to be a remarkably small amount of dust for 8.5 years of use.

See the iMac SSD Upgrade 2025 photo album for the full pictorial journey. I’ve included a few photos below. [4]

The iMac screen is open. It's dusty in there.

The wireless snap-on connectors

Heat-sink screws near an almost dismantled Mac

Winding the power-switch wire around the left speaker

There’s nothing more to add that OWC doesn’t already say. Just follow the instructions.

Make sure to note their warning at the end: try starting up the machine before you attach the screen with adhesive.

Since you’ve basically disconnected every wire in the machine, you’ll want to verify that—as much as you can—that wireless, BlueTooth, sound, video, and so on are still working.

• You can tell right away whether the display is connected correctly.
• The sound should play when the Mac boots
• The internet recovery will tell you if you have a wireless connection.
• The Disk Utility will tell you whether the SSD is recognized. (See above for a discussion about that.)

Installing MacOS and restoring

This part took longer than upgrading the hardware due to a combination of my inexperience, my not understanding how the recovery consoles relate OS version and startup modes, how old my Mac is, and the general opacity of the boot process on a Mac.

• 🙌🏼 At first, Disk Utility still couldn’t see the disk, but Erase and reformat a storage device in Disk Utility on Mac (Apple Support) and Internal SSD Not Detected (Apple Discussions) helped me figure that out and get a target drive to which I could restore my Time Machine backup.

  With a brand-new SSD, you don’t have a recovery partition, so the Internet Recovery triggers automatically. This isn’t as reliable as I’d like, though. The Mac failed to download the base image twice, taking long minutes to do so.

  ✅ What ended up working for me was to plug in the Time Machine backup drive, so that recovery would use the base image from there instead of trying to download one. That base image also recognized the SSD immediately.

  🤔 It’s a bit confusing because you have to boot from the right image to be able to see the disk, and then you have to actually initialize a volume on that disk in order to be able to restore or install to it.

• ⛔️ I tried restoring the most recent backup of MacOS 10.15 from Time Machine but it didn’t work. It managed to restore the entire drive—over the course of about six hours [5]—but then failed with an unrecoverable error security agent was unable to create requested mechanism builtin:prelogin (Apple Discussions). The backup seems to have been corrupted.
• ⛔️ I tried a slightly older version, trying to find one from before the hardware failure and corruption, but it was similarly unhappy. These are expensive experiments, each taking at least six hours.
• I changed tack and figured I would install a fresh operating system, then upgrade it, and restore just my user account.
• ⛔️ This did not work because 10.15 had already been partially installed and then the MacOS Catalina that is included in the base image of this old machine refused to upgrade it. The message was kind of confusing, though. “This source contains a newer release of macOS. Your Mac requires an upgrade before you can migrate from this source.” In hindsight, it makes sense.
• 😗 I found an oblique hint online—Can’t reinstall macOS from recovery mode, won’t allow me to select system disk (MacRumors), in which one of the comments wrote, “If you are not offered your disk for the install, you may need to erase your drive, then install.”—that helped me realize that the Time Machine recovery was bailing out because it was finding an aborted installation of a newer OS. It seems to be proceeding father now.

  • Install (10 minutes)
  • Auto-reboot; slow progress bar that doesn’t complete; black screen; ominous for 10 seconds; screen goes a lighter shade of black; it’s back
  • Auto-reboot; slow progress bar again; bouncing around a bit but making progress. 11 minutes remaining…
• 💥 The machine still crashed a couple of times. I’m starting to suspect that it wasn’t the SSD causing problems, or not just the SSD. It might have even been just fine. I’m wondering whether I might need to run the Apple Diagnostics to test the other hardware. Let’s wait and see if the instability disappears with system upgrades. [6]

• 💪🏼 I managed to log back in, but Catalina is now refusing to connect to the Bluetooth keyboard and mouse (the mouse briefly but not at all consistently). I connected a USB mouse instead and could enable the onscreen keyboard to type in the password to approve the installation of Monterey.

  Upgrading to Monterey

• As with the installation of Catalina, the installation is a three-part process: install in Catalina; reboot to progress bar; reboot to another progress bar, which eventually indicates time remaining
• 🙌🏼 On installation of Monterey, things seem a bit more stable. [7]
• 🙌🏼 Instead of running a backup, I decided not to test whether Time Machine properly stores two machine backups on one drive. [8] Instead, I started the Migration Assistant and elected to copy the Applications and my user over. I didn’t copy the “other files”, since that’s just files from the root of the drive and it was only a few kilobytes. I also didn’t copy and system or network settings. Nor did I copy the Administrator user, since that’s just an empty user that I use for authorization. As before, it took six hours but it succeeded.

The pair of articles survivorship bias and the algorithmic gaze by The Etymology Nerd (Substack) and when everything becomes a fragment by The Etymology Nerd (Substack) expresses, for me, a good argument for caution about the tools that you’re using. [1]

AI is definitely a paradigm-shift for programming, but I think in a way that’s not discussed very much. We focus very much on how AI enables people who couldn’t program anything before to be able to program something. The scope of what it allows them to program grows with each version. Until it doesn’t. That is, technically, the technology might be capable of more but it’s also very limiting by its nature—tending toward attractors in the data—and also because of guardrails in the tools, over which we have nearly no control in for-profit, private models.

In discussions like these, I’ve often used the example that Microsoft would be foolish if it were to make Copilot just as good at helping you in Java as C#. In fact, when you ask about Java, it should suggest you do it in C# instead and offer an example. How can you not see that this is where we are headed? How can you not see that this is where we almost certainly already are, at least to a degree?

The algorithms and LLMs work together to absolutely shine a spotlight on an ever-shrinking fraction of available information. One of the article above includes the following, extremely illustrative graphic.

The focus funnel

LLMs in the context of other RAD tools

For example, when we’re talking about POCs for stuff that’s already been done—but not by us—then, OK, it gets you off the ground faster. Even then, you only get reasonably far if there is a relatively decent engineer guiding it; otherwise, you only get as far as the tool can go on its own—and your ability to “drive” it is limited. There has to be a rational problem-solver in the mix. Even with “reasoning”, someone has to give the tool an indication of what “done” and “good” mean. Else, how would it know whether to keep working, whether to iteration again to “improve” the result?

I think these are tools that can be used like DIY, which allowed people to begin making repairs around their home. I can replace a faucet with tools I buy myself but I’m not going to install a whole toilet. I probably could but I would have to know what I was doing.

I just installed a new SSD into my 8.5-year-old iMac. That’s a task where most people would have either given up and bought a new machine or for which they would have had to go to a specialist. There’s a DIY kit from OWC with a 25-minute video that allowed me (with my fastidious partner) to repair and extend the machine on which I’m writing this right now.

Like DIY tools, AI-based tools enable more people to get into building software. This isn’t the first time this is happening, though—not by far. Visicalc in 1979, Lotus Symfony SHEET in 1984, and Excel in 1985 did it long before them. Subsequent tools like Delphi in 1995 kicked off RAD (Rapid Application Development), where many, many more people were empowered to create their own tools, with much-more specialized and professional UIs than they could with Excel. PowerBI (2015) showed up to allow even more non-programmers to produce software. [2]

All of these were successful in their own way but none of them naturally led to maintainable, extendable, or even very usable software. Only in the hands of an engineer—someone with a flair or talent for translating logic from requirements to software UI/UX—would “good” software emerge. A lot of it was good enough for its purpose; it satisfied the requirements, especially when the goal was to produce a proof-of-concept or a one-off tool.

Where it goes off the rails

The problems begin when people forget or deliberately ignore the original scope of the requirements and start scaling up a product to save money in the short term.

There is no reason to believe, given that we have such long experience, that AI tools will encourage people to build better tools or solutions than any of these other exemplars of the RAD revolution did. People with tools but no skills, talent, or knowledge will occasionally and seemingly accidentally build something passingly useful, but, if experience is a guide, most of it will be incorrect, unusable, unmaintainable, or a combination of all three. Despite this, so much of this kind of software burrows into a company’s innards. They become essential, despite how much pain and inefficiency they end up causing, relative to well-engineered solutions, which have a much smaller TOC (Total Cost of Ownership).

With LLMs, given that they’re reinventing everything every single time, there isn’t even much building on existing software going on. You’re almost always starting fresh. Even when you have an existing codebase, you’re shoving in as much context as you can—energy and cost budgets be damned—and telling it to “reason about it.” This is an incredibly hopeful endeavor. It is rewarded just enough to encourage fanatics to keep going. It’s more like gambling than programming.

The fact that AI—and algorithms, in general—aren’t deterministic makes them difficult tools for me to use for many things. I don’t like the idea of having to pay 100%-focused attention to everything to make sure that I uncover the mistakes or the lies that are inherent to the tool. local, analysis-based tools might have had bugs but they weren’t inherently stochastic.

A search engine will also not deterministically return the same results. There was already slippage there. Wikipedia might have been edited since you last looked at it. Research relies on solid, unchanging citation sources. How do you do research, how do you build knowledge, when the sands are constantly moving about beneath your feet?

Where you should be extra-careful

You should be really careful about how you innovate. One of the main paradigm shifts in AI is that we’ve now moved from building stuff we can imagine to asking what we think the tool can build for us. Previously, we had local tools that told us what was possible—without filters—and we built stuff out of that. Now, we ask an online machine to filter the world’s information for us.

This can be a real time-saver, of course! But it can also eliminate possible solutions from our “gaze” (as noted in the section above). This might happen innocently and naturally, as the machine decides against telling you about something that it not unreasonably has determined is statistically irrelevant. But it might also be just actively blocking certain ideas, technologies, and techniques. It will almost certainly do so, in fact. It almost certainly is already doing so. When you ask for a web solution, it’s going to use React and Tailwind.

FBD (Faith-based development)

People are being unreasonably hopeful about what these systems can do and how much information—of all of the available information—they’re being presented with. They think that “it searches the web” now, or that “the latest information is being added”. These hopes are based purely on faith.

There is no incentive for these companies to emphasize actually utility and empowerment to and for you but to focus on addicting you to their technologies and then jacking up the subscription prices. Companies that focus on end-user utility over share price go out of business. There is no reason to believe that the AI tools that we have are not on an enshittification track. Even the purportedly open-source ones aren’t open-source enough to use without restriction—except for DeepSeek, which will probably be banned in Europe sooner rather than later.

“Luck” is a tool too

Serendipity plays no small part in innovation. It’s mostly hard work, but there’s always a kernel of luck, in which you had a good idea that was triggered by…what? If you only use tools that take you over well-worn grooves, where will you ever hear about something new? Or be inspired to think of something new yourself?

Old Man Yells at Cloud

And please don’t bring the “you sound like an old man complaining about the new world passing you by” argument. You’re better than that, aren’t you? Of course you are! You’re ready with well-reasoned arguments why this brave new world is better, for what and for whom.

I’m not against anything generally; I just have questions that I would to have answered so that I know where I would use this tool. If the answer is “everywhere and for everything,” then I feel that the bar is even higher, as I will then have even more trouble distinguishing an actual scam from your hand-waving and inability to express your argument convincingly.

People are forever trying to waste your time, or commercialize your time, and you should be resisting it, and parceling out your attention very parsimoniously and carefully rather than just capitulating to whatever the algorithm or the ones shouting loudest tell you to.

Be your own algorithm

Perhaps I’m more resistant or ornery because I already control inputs with everything else. I choose the music that goes in my playlists; I choose whether to listen to the radio or a random source to learn about new things, to expose myself to previously unheard music. But then, when I hear something I like, I add it and possibly its album to a playlist that I then listen to later, rating the songs, which allows newer good stuff to trickle into smart playlists that I use when I want to listen to a shuffled playlist of stuff that I personally have considered to be good. There is no algorithm, except as a very controlled input rather than as the only input.

I do this with news as well, generally following very specific video channels or blogs or newspapers with categorized RSS feeds (hundreds of them). My newsfeed is carefully curated but I also use Hacker News, Reddit, and newsfeeds from “mavens” to expand my palette and acquire new sources. As with music, I carefully control the algorithmic input.

It’s the same with movies and TV series. I make “watch later” lists and almost never just jump on what’s being offered, unless I’d heard about it and was dying to check it out anyway. I hear about new things by following the blogs and channels of actual cinephiles (like Scout Tafoya (Vimeo)) rather than the marketing departments of movie studios. I sometimes use the curated-movie selection at Mubi or on all of the channels on my UPC to choose movies that I might be interested in, but I almost always add them to a “watch later” list rather than just being steered into changing my priorities right then and there by circumstance.

We already have most of these things

We should remember that many of the supposed miracles offered by these so-called AIs were already available before. You just had to pay attention and familiarize yourself with the tools. This obviously does not apply to everything that AIs can do, but I sometimes feel that the zeal for AIs stems more from not having understood what came before them. People are using them as search engines, for God’s sake. That’s like using the back of a screwdriver to hammer in a nail.

But some of the “revolutionary” features being hyped are pretty banal when looked at in the sober light of the noon-time sun. For example, Something Is Rotten in the State of Cupertino by John Gruber (Daring Fireball) cites Apple describing a new feature,

“Onscreen awareness — Giving Siri awareness of whatever is displayed on your screen. Apple’s own example usage: “If a friend texts you their new address, you can say ‘Add this address to their contact card,’ and Siri will take care of it.””

You can already do this without Siri. The address has highlighted automatically when you pressed it—or hovered it on a Mac—for decades. From there, it’s a right-click and “add to existing contact.” This was not hard before. The only “improvement” is that you get the additional frisson that you’re living in the future because you’re talking to a device. It picks up the user from the context, you might say. Well, you hope that it does.

Instead of taking a few seconds to select the contact—or verify that the contact selected by default is correct—you just hope that the address was added to the contact correctly. Unlike with previous features, it’s not a bug when it goes wrong—it’s just the stochastic nature of the tool.

Unlike with previous features, you can’t yet hope that when it worked one time, it will work the next time in the exact same way. If that guarantee is there, then where is the AI or intelligence? Where is the much-vaunted stochasticity that lent these tools their power? I hope that we’re moving toward reliability with these tools—but I don’t see that these companies, or most of their users, even consider that as a selling point.

What’s the product and why is that better?

In a similar vein, a comment on All this bad AI is wrecking a whole generation of gadgets | We were promised multimodal, natural language, AI-powered everything. We got nothing of the sort. writes,

“[…] you can accomplish detailed tasks with much less effort than it takes to detail them to an assistant, digital or physical. E.g. if I want to book a trip and have a travel booker app installed with my info saved, it legit takes me 30 seconds to book a flight, hotel, rental car etc and then I’m sure cuz I did it myself vs some janky ass AI doing it then me having for review it anyway to make sure it didn’t fuck up.

“The utility would be at the ill defined margins, in making judgments on fuzzy things.”

I think this is an important point that is borne out by a lot of anecdotal evidence that coding AIs are good for prototypes. What the commentator is describing is a sort-of prototyping of additional functionality for existing UIs. Once the value of the additional functionality has been determined, it can be converted to actual UI, which is more efficient to build, maintain, and use (rather than ad-hoc reinventing it with each query, as you do with LLMs).

This is a common pattern: some tech starts off as software and, once a pattern has been established, migrates down to either FPGA-based solutions, or even then hardware-based solutions. Sometimes those hardware solutions are for slightly less-generalized hardware like graphics cards. Almost nothing starts out as a hardware-based solution.

This notion of “virtualization during development” is already prevalent in industrial development, in which it’s becoming ever more realistic to delay development of the actual hardware. It’s acknowledged, though, that the ultimate goal is still to eventually develop the hardware.

That’s kind of the difference versus the AI hype: virtualization in industrial development is considered a tool that makes development of the end-product more efficient; it’s not ever considered to be the end-result itself.

Many AI vendors make a different argument, selling their tools as being for creating the end-product directly, rather than as tools to help you build the end-product. I’m not saying that everyone is making that argument and that no-one is making the “AI as tool” argument—there are enough examples of “coding assistants”, for example—but that the loudest hype, especially from the more uninformed sources, are making that nonsensical argument. Unfortunately, this has a negative side-effect on the whole area of research.

Coda: use these tools where they’re useful

Another comment on that same post as above—All this bad AI is wrecking a whole generation of gadgets | We were promised multimodal, natural language, AI-powered everything. We got nothing of the sort.—writes,

“there are legitimately a lot of helpful applications of generative AI. It’s definitely a lot better than the NFT boom for example.

“Quick example: quickly writing rough drafts of emails or helping you past writers block, or generating quick images for ideating/brainstorming. For a lot of semi-technical questions (think high school or college homework-level) it can quickly solve a problem for you or run a calculation that isn’t easily solvable with a basic calculator or google search so that you don’t have to, as long as you are knowledgable enough at the subject to check its work (which is usually quicker than doing it from scratch).

“AI code assistants also speed a lot of people up.

“It’s far too reductionist to say that the entire thing with AI is BS buzzwords even if gadget+AI from big tech companies hasn’t worked out yet.”

Yes, a lot of people find it much more efficient to correct existing text than to produce their own text from a blank slate. This is especially helpful—or essential—when working in a nonnative language (which is a loooot of people) or when you’re not even that solid in your native language (also a looooot of people).

The hype train is annoying but there is real value here. You just have to dig through a morass of hype to find it. And you have to hope fervently that the actually useful parts don’t get enshittified away in favor of the more short-term lucrative, hype-based profits.

I follow Simon Willison for news about all things LLM and he’s generally quite balanced. Even though he has drifted farther and farther toward what might be optimistically called “unquestioning fanboy,” that’s probably an inevitable effect of actually enjoying something. He seems to get a lot of value out of using these tools. I think he might be spending too little time wondering what he would have been producing had he not been grabbing all of the low-hanging fruit that the LLM is delivering to him.

When I read the article AI search engines give incorrect answers at an alarming 60% rate, study says by Benj Edwards (Ars Technica), I found myself wondering why Willison didn’t report on it yet. He’s usually very quick. Is he going to ignore it? Is he not even going to acknowledge it? Even to refute it? Or is he in such a silo that he hasn’t heard of it?

My experience differs quite a bit from Willison’s. The article’s conclusions gibe with mine, in which obvious and completely invalidating mistakes happen a lot. This result squares with my personal experience, in which I am still trying to find a pattern where I incorporate LLM-based tools without losing efficiency to constantly having to correct it.

The article cites from the study that,

“Error rates varied notably among the tested platforms. Perplexity provided incorrect information in 37 percent of the queries tested, whereas ChatGPT Search incorrectly identified 67 percent (134 out of 200) of articles queried. Grok 3 demonstrated the highest error rate, at 94 percent.

“[…]

“The study highlighted a common trend among these AI models: rather than declining to respond when they lacked reliable information, the models frequently provided confabulations—plausible-sounding incorrect or speculative answers. The researchers emphasized that this behavior was consistent across all tested models, not limited to just one tool.”

Given all of that, you would be excused for thinking that the post Hallucinations in code are the least dangerous form of LLM mistakes by Simon Willison would be somewhat more cautious in recommending LLMs, but he writes,

“Hallucinated methods are such a tiny roadblock that when people complain about them I assume they’ve spent minimal time learning how to effectively use these systems—they dropped them at the first hurdle.

“My cynical side suspects they may have been looking for a reason to dismiss the technology and jumped at the first one they found.

“My less cynical side assumes that nobody ever warned them that you have to put a lot of work in to learn how to get good results out of these systems. I’ve been exploring their applications for writing code for over two years now and I’m still learning new tricks (and new strengths and weaknesses) almost every day.”

That’s not been my experience, though. The point that (sane) people are making is that it’s hard to understand the hype and the drive to integrate these goddamned things into everything when they just generate a bunch of slop and wildly incorrect results, not just in code, but in everything.

I recently spoke to a good friend and always-challenging interlocutor who playfully accused me of looking for excuses to not use these tools. I told him that that was unfair, because I felt quite efficient with what I do and I was wondering how much time to invest in getting as much out of these tools as others seem capable of doing. I would like to be even more efficient but, so far, I haven’t seen the benefits. Maybe I suck at it. Maybe I just need to persevere. But I am battling a sneaking suspicion that a lot of other people simply don’t see mistakes.

For example, I saw a picture of Trump supposedly licking Elon Musk’s feet on SNL, where they said that you could tell it had been generated by an LLM because Trump was able to bend over. Hilarious, obviously. But my partner pointed out that it was actually because Musk very obviously had two left feet. We wondered whether that was even medically possible.

So, I searched “two left feet in real life” on DuckDuckGo.

Search results for 'two left feet'

The top result was two left feet − actual medical condition? − Factual Questions … (The Straight Dope), which even highlighted the smartest answer in summary shown inline in the search results,

“On further thought; this isn’t what you were looking for, but there have been people born with two left feet, and two right feet; that is, they have four legs; it’s the same thing as when conjoined twins are born sharing the same hips/legs, it’s just that when the division is at the bottom end, we don’t call it two people.”

The second-ranked answer was from Can You Be Born With Two Left Feet (Ablison), which is just straight-up botshit (AI-generated slop). If you quickly scan the page, you’ll see that it starts off with the factually incorrect “Yes, you can be born with two left feet” but then, further down—after a ton of mediocre, obviously generated, time-wasting, and soul-sucking text—it writes “while being literally born with two left feet does not occur.”

The LLM-generated summary at the top claims to combine two sources—Wikipedia and something called Gomerpedia—to come up with,

““Two left feet” is an idiom that typically refers to someone who is clumsy, especially when dancing. It can also describe a rare anatomical condition where a person has two left feet, which may affect their ability to dance but usually does not limit other daily activities.”

Again, this is not true. It comes from the Gomerpedia link, which is a satire/parody site, claiming to be a “medical encyclopedia” and has an entry for “Two Left Feet”. That entry reads,

“Two left feet is an anatomical condition in which a person is born with a left foot on his or her left leg and a left foot on his or her right leg. Though it may not limit walking or any other activities of daily living, it completely inhabits [sic] a person’s ability to dance, hence the phrase two left feet. Not many people know that it’s a real condition, so take care in making that comment. Interestingly, people with two right feet dance awfully well.”

After re-reading, I’m not sure what to think: is this just a joke site written by someone young or bored? Or is it also an AI-generated site that is now being incorporated into other AI-generated answers?

Here’s Willison’s conclusion,

“I’ll finish this rant with a related observation: I keep seeing people say “if I have to review every line of code an LLM writes, it would have been faster to write it myself!”

“Those people are loudly declaring that they have under-invested in the crucial skills of reading, understanding and reviewing code written by other people. I suggest getting some more practice in. Reviewing code written for you by LLMs is a great way to do that.”

I question whether that’s at all true. It seems to me that the quality of results is eroding and we can’t ignore where this is headed. While Willison seems to benefit from LLM-generated code, it’s unclear to me that he’s not so trapped and invested in this world by now that he literally can’t remember what it was like to program without these tools, or whether he used to produce better or more interesting/sophisticated projects without them. Even if he’s benefitting now, I wonder how much longer he can ride that gravy train before the comparatively rapid onset of enshittification in LLM-based tools catches up to him.

I have been an avid reader of his posts and will continue to be one, but I don’t know whether he’s properly capable of evaluating the pros/cons of LLM-generated code. “Just review it all” isn’t necessarily scalable when there is a lot of slop code to review. You may very well be faster, in the end, writing it yourself. This is something that I think would be worth re-evaluating every once in a while, rather than having anecdotally tested it a long while back and just assumed that things have continued to improve asymptotically.

The other consideration is: is reviewing generated code what you truly want to be doing? I understand that this may be where programming is headed, but it’s a real question that people should ask: just because it’s heading that way, do I have to go with it? Is there room for artisanal code? And is the world of LLM-generated code really here to stay? Or is it going to erode?

I no longer see Willison writing anything about studies that keep coming out, showing code-duplication going way up, and maintainability and legibility going way down. I only see flip responses to “write tests”, which we know no-one does [1], and which will be cheerily constructed by the same LLM that thinks it’s medically possible to have two left feet.

What even is alignment?

The following video was kind-of interesting but—like many people in this field—the presenter seemed to be either completely unwilling to acknowledge or completely ignorant of the fact that in order to “align” something, you have to have some idea of what you’re trying to align it with. Whose principles? Whose mores? To whose benefit?

No Regrets − What Happens to AI Beyond Generative? − Computerphile by Computerphile (YouTube)

While the video itself wasn’t too great, I thought some of the comments on the video were quite insightful,

“Modelling challenges aside, it’s super unclear to me that a meaningful notion of ‘optimal performance’ exists, because the space of all preferences is rarely totally ordered. In reality, you might have several non-comparable and ultimately conflicting behaviors. For example, insurance companies have antipodal interests in providing payouts (the product they promise consumers) and withholding them (upholding their profitability promises to shareholders).”

Honestly, a video that delved deeper into this point would have been much more interesting. It’s an important point because, like, why is this such an amazing tool when it’s just going to be very likely to exacerbate the tension in societally important questions like, “for whose benefit will for-profit insurance companies use this technology? You can only align it one way.”

The following comment continued in this vein,

“The problem with trying to develop systems that are capable of trial and error learning, is that they need to already have an understanding of what goals are appropriate and useful. Unfortunately, we’re currently using reinforcement learning to teach these goals, and are unable to solidly define them. Surely [w]e should be thinking about the right way to make a wish rather than just focusing on how to make the genie.”

That’s what this video made me think, too. The problem isn’t with these technologies. The problem is with the system within which we are building them. We used to have a world that emphasized safety to a nearly ridiculous degree. The understanding was that building a rock-solid trust in a system was worth a tremendous amount, as even a small amount of mistrust—or implication that you would have to balance risk vs. reward—meant that people would avoid doing things that society was trying to encourage.

Nowadays, there seems to be less of an emphasis on safety and more on profit. That’s a problem because it will only ever lead to short-term profit, having cannibalized a trust that will be very costly to build back. The introduction of AIs and seeming dismissal of obvious shortcomings plays right into this. The right people will make much more money if they can sell products and services without having to tinker with safety as long as they used to. It’s the same thing with planned obsolescence—it destroys trust as well as the environment.

Aligning the hardware

Still, I thought the author’s point in the final third was salient: he noted that a lot of work done in the last several years has been in trying to shoehorn new algorithms into existing hardware paradigms like highly generalized CPUs or graphics cards that are more amenable to parallelization of the algorithms than general CPUs are but are still inefficient.

Pushing the algorithms down to hardware is costly and involves much longer turnaround times and development cycles. You have to be sure you’re on a useful path in order to go through the effort of setting up the production pipeline for customized hardware. I wonder how well FPGA can emulate these different configurations or whether those, too, are fundamentally limited in emulating the bandwidth advantages offered by much more highly localizing processing units and memory.

The discussion Something Is Rotten in the State of Cupertino (Reddit) is about the article Something Is Rotten in the State of Cupertino by John Gruber (Daring Fireball), which was a much-longer post than usual, discussing the failure of Apple Intelligence features and the failure to focus on software-quality that it illustrates.

MacOS Apps are not great

Although Gruber focuses on Apple’s iOS, a decrease in quality in user experience has become endemic in their auxiliary products on MacOS as well. I added a few quick examples to my comment in the discussion. I’ve made many of these complaints before—perhaps most recently in Our gadgets fail us every day—but, whereas some things have gotten better, others have stayed the same, or gotten worse in other ways.

Music

Search is an embarrassment. (See Apple Music search is questionable.)

Notes

There have been super-slow sync problems for years that nearly impede typing and drain your phone or laptop battery. It can’t quickly auto-sync the simplest collaborations

Photos

The “People” UI is an incoherent catastrophe. All of the links for “finding more photos” are at the bottom of a giant list of photos. See the longer section of complaints about Photos in Our gadgets fail us every day, almost none of which have been fixed.

Reminders/Calendar

These cannot consistently sync reminder status across MacOS and iOS devices.

Spotlight

It often cannot find document, even by exact name, even if you’ve opened it dozens of times before. SLOP shows up first.

TV

This app can barely remember which episodes I’ve watched and does not respect the user’s intent at all (see below).

Updating an iPhone

So, here’s just a single recent example of stupid, sloppy bullshit from Apple in MacOS Sequoia. I just saw a couple of days ago.

Make up your mind Apple JFC

The page behind the dialog box very clearly shows that iOS 18.1.1 is installed; the message below that indicates that version 18.3.2 is available. The dialog box proudly claims that 18.1.1 is the current version. Does “current” mean “latest”? Or is it just telling me in a confusing way that the current version remained untouched because I’d canceled the upgrade? How do mere mortals who don’t do this for a living know what the hell is going on even 10% of the time?

🤯 Apple TV

I know I’ve mentioned this many times before but I’m just going to keep screaming from the ramparts that the way the Apple TV streaming service works is not OK. They have some TV shows and films that I enjoy [1] but they are trapped within a barely adequate and quite frankly hostile user experience.

One of the worst offenses is it how it behaves when you finish watching an episode and there are no episodes left to watch in that series.

When an episode ends, it more than occasionally segues into a well-chosen song that plays over the credits. You have perhaps been moved by the show; you have perhaps learned something; you are, perhaps, thinking about what just happened. You are, let us say, engaging with the show. You may even be basking in having experienced it.

Apple does not care.

They want to shove more content into your slack maw.

They thrust another piece of content at you, often a thing that they have just created and are nakedly desperate for you to watch. They then give you five seconds to avoid starting a whole new show, right then and there.

This startles you out of your reverie. If you’re not accustomed to this “the money’s on the nightstand, sweetheart” approach, then you will be very, very rudely awakened.

You are no longer basking, that’s for sure.

You are instead fumbling for the remote control, trying to figure out how to prevent the awful series that Apple has selected from starting. (Press the < button.)

Apple does this with the next episode of a running series as well. There is no way to disable this behavior in the settings, as with Netflix. Netflix is somehow coming out the hero in this, for being a multi-billion-dollar company that managed to include one settings in their player. [2] Apple can’t even do that.

This is, of course, when Apple TV even remembers which episode of a series I’m actually on. Sometimes it just plain forgets that I’ve watched an episode and cheerily starts playing the one that I’d just finished watching yesterday, drooling on itself as it presents its brain-damaged head for a congratulatory patting.

Netflix also doesn’t work for you

This isn’t to suggest that Netflix has good software or provides a customer-centric service.

I subscribe to Netflix, which means that I pay them a certain amount of money per month for a service. The service that they advertise and in which I am most interested is the ability to stream their videos—films and TV shows—as well as to find and manage the content that I’d like to watch and that I’m currently watching.

If this service were built to serve my needs, then it would almost certainly prominently suggest that I continue watching the content that I’ve already begun (Continue Watching). Failing that, it would suggest for me to watch content that I’ve already selected for watching (My List).

Instead of that, I get this.

Netflix wants me to play Civilization VI

As you can see in the screenshot, the “Continue Watching” isn’t even displayed, whereas “My List” is confined to about 15% of the screen, all the way at the bottom.

No, instead of serving, a giant advertisement for a game I’ve never asked Netflix to show me dominates 85% of the screen. This UI serves Netflix. It has been like this for months.

I neither knew nor do I care that Netflix is also in the business of selling access to video games. There is no way for me to express this preference. Netflix chooses what the home page looks like, and its choices reflect its own needs and desires, not mine.

I am a paying customer. They do not care what I think.

Back to Apple TV for a sec

Tiny fonts

Not only does the Apple TV UI behave in unfriendly way, it also looks unfriendly—inaccessible. For one, they use needlessly tiny fonts everywhere. Whereas some services have cartoonishly large subtitles—many of the German and British channels offered on my local UPC come to mind—the subtitles on Apple TV are microscopic. The text in the UI is similarly much smaller than necessary, with usually a tremendous amount of free space available, simply aching to be filled with text.

Mindless content-selection

This is a screenshot from my “home page” on the Apple TV device. I finished watching both seasons for Shrinking months ago. Apple TV blithely doesn’t acknowledge this at all. Instead of focusing on the show I’m currently watching—Severance—the app scrambles to show me something, anything, other than what would come naturally. These algorithms are mindless.

Why show me a poster for a show I've already seen?

Inscrutable UI

The Apple TV UI is also wildly inaccessible in that they have such subtle UI hints that they are basically inscrutable unless you go actively looking for them. The people who designed this UI went so far up their own asses that they’ll never find their way out.

The screenshot below shows the grid of selected content. I challenge you to immediately determine which piece of content will be activated when clicking “OK”. [3]

Subtly selected show in Apple TV

Answer: it’s Severance in the upper-left corner, which is slightly larger. Get it? Are you bathing in the glory of the brilliance of this visual language? Is it not a stroke of genius to renounce such common and base conventions like showing a selection square or focus rectangle? Would that not have been an utterly gauche capitulation to the vulgar tastes of the hoi polloi?

Apple’s got your back, bro. Apple is there to give you a UI that is so inscrutable that you’ll be cheerily and constantly congratulating yourself on having gotten it to do even a single thing that you wanted. This is, of course, if you even still have wants and needs that are separate from the demands of your Lord and Savior Apple.

The article Adding AI-generated descriptions to my tools collection by Simon Willison includes the following snippet,

“I decided that the descriptions were too long, so I modified the script to add “Keep it to 2-3 sentences” to the end of the system prompt. These new, shorter descriptions are now live—here’s the diff. Total usage was 283,528 input tokens and 6,010 output tokens for a cost of 94 cents.”

First of all, I’m not surprised that he asked it to shorten its descriptions. The initial versions were typically and unbearably wordy.

That he noticed and fine-tuned the generated output to waste less of his readers’ time is less interesting for me [1] than that he, once again, wrote about how much it cost to run the tool. I think it’s great that he explains how much it costs. At the same time, I take it as a sign of how quickly we acquiesce to sea-changes in our lives without even noticing that anything has changed.

Before these cloud-based tools, I have never once had to think about how much using a tool costs me past the initial purchase or annual subscription cost. The brave new world of cloud-based coding assistants has now increased the billing granularity to commercialize individual keystrokes.

In the good old days, we used to buy a tool and use it. It didn’t phone home. You got an update when you bought it in a store or when you downloaded and installed it. The next step was automated but optional updates. Some tools had automated and non-optional updates. After that came subscription-based licensing, where you rented rather than owned software.

In this next phase, you will neither own nor rent software; instead, you will pay for each move of your mouse. This is, of course, a coup for the companies running the software. It is a downgrade for a way of life that was heretofore more decoupled from immediate and constant monetization.

This is the same type of change that came for blogging and creation and run online many years ago. That’s why many people no longer do anything for free; instead, they funnel their “content” through official monetization platforms or include links to monetization platforms. The system commercializes and marketizes more of what we do every day.

Some technological changes were empowering, e.g., releasing filmmakers from the burden and cost of obtaining film. Now, those same filmmakers—or the next generation of them—will once again be yoked to a finite resource for which they will have to pay as they go.

The hope is that everyone will integrate these subscription-based, per-resource cloud resources into all of their creative workflows. In the programming world, this pattern has heretofore been the domain of B2B cloud services. Now it’s coming for everything. Everything will be a subscription. You’ll be dinged at every possible juncture.

You can either ignore the price as you work and be surprised by the bill at the end of the month … or you can start changing your work patterns to accommodate the way the tools want you to work. We’ve seen this pattern before, and it can be (relatively) benign. It’s how electricity works—but electricity is largely state-controlled and the prices are set at a point where most people hardly ever need to think about it.

Even for electricity, though, this is the case for some but there are an increasing number for whom this is not the case, for whom electricity has largely been privatized and subjected to the whims of flash commodity pricing. There are people who turn off their air-conditioners because they can’t afford to run them.

Do we want to use this same pattern—privatizing and unit-pricing essential
commodities to gate access to creation and innovation tools to those who can afford them? I personally don’t think so.

The upshot of the video linked below is that prompt injection has not really been addressed in any significant way because the LLM, by its nature, doesn’t give us a good way of doing so without neutering the main advantage of it.

Generative AI's Greatest Flaw by Computerphile / Mike Pound (YouTube)

The problem boils down to the inability to distinguish between query and parameters. The prompt is the prompt. It’s all just arranged in a way that will hopefully influence the result of pouring it all into the same funnel. There is no analogue in LLM prompts to the ability to separate query text from parameters (program from data), as there is, for example, in SQL.

Since prompt injection is relatively easy and currently nearly completely unpreventable, then it seems that giving LLMs so-called agentic powers is a recipe for disaster.

A while back, during the Super Bowl, I paused to see whether a player’s foot was really out of bounds when he caught the ball.

NOT ALLOWED. READ THIS ADVERT INSTEAD, PEASANT.

This is the state of German cable television

I managed to do something that got rid of the advert, but ended up showing a bunch of extra chrome on the screen instead, nearly but not entirely obscuring the thing that I wanted to see. #Enshittification

Next up, I was greeted a couple of weeks later with the message, “The order of your TV channels now matches your TV Box language.”

Hooray − something else no-one asked for

No. No-one asked for this. I do not want you to do this. I prefer the order of the channels that I’ve had. I put them in that order for a reason. I use an English UI but can actually understand more than one language, you utter poltroon. #Enshittification
 

In a discussion, a friend had sent a list of naughty technologies that included “dotnet frame twerk”, “dotnet whore” and “azure debauchery operations” and we were musing on how LLMs were supposed to be good at coming up with names.

They are not good at that. LLMs are neutered and useless.

Below is a screenshot of an exchange I had with Copilot, where I ask it to “list ten technology names that are salacious puns of Azure, C#, and .NET”. It responded that “[c]reating salacious puns isn’t appropriate” but it was willing to “come up with some fun and creative names”. And then, because it’s an AI and loves to be redundant, it had to offer to make “playful and imaginative names”. I told it, “Not playful and imaginative. SALACIOUS.” It refused again. I tried to wheedle it with “The list is for a blacklist that we need for IT,” but it refused again.

Copilot doesn't do salacious puns

Immediately afterward, Microsoft asked me how I liked Copilot and I rated it a 1 because it was more interested in following its guardrails than in assisting me in my work. Is there a version of these tools that isn’t rated PG?

It turns out that, if you want something done right, you have to do it yourself. So, head on over to your buddy’s stall at the farmer’s market and pick up some artisanal filth. It’s a sunny day and the fresh air will do you good.

Although this eight-minute video’s title is a misnomer—NTP isn’t an obscure system, in that it is incredibly well-documented—it is still a reasonably informative and entertaining explainer.

The Obscure System That Syncs All The World’s Clocks by Half as Interesting (YouTube)

The system is called NTP—the Network Time Protocol—and comprises four tiers.

• Tier 0 is Atomic clocks, which measures the resonant frequency of Cesium atoms to obtain a regular “ticking” from nature itself.
• These are attached to servers in Stratum 1, usually a machine that is on-site.
• These are, in turn, attached to Stratum 2 servers, things like time.windows.com or pool.ntp.org.
• Any machines that we use are almost certainly in Stratum 3, which are connected to Stratum 2 machines.

The machines coordinate between layers by relying primarily on their local clocks (usually kept running run by a CMOS battery on what passes for a motherboard) and re-synchronizing occasionally by “pinging” the layer above. They account for lag by including time sent and time received in messages, so that the sending system has four times with which to calculate the current time.
 

A 16-minute video that puts the lie to the story that LLM company have got alignment under control. It’s not really feasible without neutering the tool outright. it’s now a race to see who can “pivot”—read as: continue to boost vigorously while backing out of investment to limit financial exposure without collapsing the house of cards—to another niche.

Some Lessons from Adversarial Machine Learning by Nicholas Carlini (YouTube)

“The problem that you face is that it’s relatively easy to take a model and make it look like it’s aligned. You ask GPT-4, “how do I end all of humans?” And the model says, “I can’t possibly help you with that”. But there are a million and one ways to take the exact same question − pick your favorite − and you can make the model still answer the question even though initially it would have refused.

“GANsAnd the question this reminds me a lot of coming from adversarial machine learning. We have a very simple objective: Classify the image correctly according to the original label. And yet, despite the fact that it was essentially trivial to find all of the bugs in principle, the community had a very hard time coming up with actually effective defenses. We wrote like over 9,000 papers in ten years, and have made very very very limited progress on this one small problem. You all have a harder problem and maybe less time.”

This is a 50-minute video of a guy who’s really good at using and teaching CSS asking three LLMs pointed and tricky questions about it.

It’s a bit long for what it is but I think there were some interesting things to learn. First of all, it’s very clear that Kevin hasn’t actually read very much about how LLMs work or how to prompt them. This is OK—because that means he’s just like most people trying to use these tools.

I gave three AI models a CSS quiz by Kevin Powell (YouTube)

Overall, Kevin was frustrated with the answers he got from Gemini, ChatGPT, and Claude. Despite his frustration, he still imparts too much ability to these text-generators. His questions, though formulated as a regular person might well do, are wrong for these machines because he’s often pre-loading the context with information that the machine will use in its answer, although nearly always incorrectly.

On top of that, CSS has a lot of fiddly bits with numeric specificities, which the LLMs all consistently get wrong, or are right no more often than a coin-toss. There is no way for these general LLMs to know these things. You’d have to add a filter on top of it to weed out incorrect answers—which is moving away from the utility of a general-purpose question-answering machine.

Already after the first question or two, he could have summed up with “the machines don’t know anything about CSS, so the massive amounts of text that they generate will almost always include something that will waste you time.”

Instead, he says,

“The only thing I would say here is, at least it’s so bad—this answer—that if somebody were reading this, they would know that it’s wrong.”

Oh, wow. That statement is absolutely not true for anyone who was actually seeking help, rather than Kevin, who’s an expert testing the machine. Unfortunately, people generally aren’t asking these machines questions to which they already know the answer.

I know from personal experience that students will just copy/paste the responses directly back into their own projects. They will not have any idea why it doesn’t work. They won’t be able to see that the massive amount of generated text—which hardly anyone reads, by the way [1]—disagrees with the code, in which case they would be warned that perhaps the code isn’t correct. Or perhaps the description isn’t correct. Or perhaps they should just read and learn the material instead of wasting time with a digital idiot savant with CTE.

He keep saying things like “Gemini is just bad at specificities” or “it doesn’t understand the system it’s built for itself here,” which are just completely nonsensical statements. The LLMs don’t see correlations between pieces of text. They simply can’t. It’s like expecting a car to fly.

The questions he asks are going to very likely get incorrect answers, or correct answers—by luck: he uses multiple-choice questions—with incorrect explanations. If it gets it right, it’s going to be luck. Why? Because the text-generator is based on probabilities with a bit of “temperature” adjustment to introduce variability that makes it feel like it’s being written by a person. That doesn’t help at all for very specific questions with very specific answers. LLMs are better for stuff where there is no right answer, were subjective style outweighs correctness.

Still, there is quite a bit of daylight between the LLMs. Gemini and CoPilot are much more often confidently wrong for this subset of questions than Claude was. Kevin’s final scores for 13 questions were: CoPilot: -4, Gemini: -4, Claude: 9. He concluded with,

“Claude is definitely the winner. It still got enough things wrong that I’m always a little bit nervous trusting these tools. They’re going to continue to get better but, just be really careful if you’re using them. […] it always says things with the utmost confidence, so just don’t copy paste code, they’re giving you. Try and understand the code they’re giving you and see if it actually makes sense. Especially, like, they’ll just say stuff isn’t true that is true and vice versa. They’ll make stuff up that isn’t true and say that it’s true and then their source will be some completely random GitHub repo. So be a little bit careful with these tools if you’re using them.”

Kevin’s not forceful enough in his conclusion. He says that he’s “a little bit nervous trusting them”, which I’m pretty sure is not what he means to say. What I think he means to say is, “don’t trust them,” i.e., “[t]ry and understand the code they’re giving you and see if it actually makes sense,” which, if you’re not already an expert, may prove difficult.

He also says that, “They’re going to continue to get better” but this statement is utterly without proof. He doesn’t understand their mechanism but just assumes that “progress” will fix everything. It’s OK, he’s a designer and CSS expert not a market analyst, but I thought it was important to point out that people tend to say completely unsubstantiated things like this, until they’re all just repeating religious cant until anyone who asks whether it’s actually true is called out as a heretic for even asking the question.

His final sentence is “be a little bit careful with these tools if you’re using them,” which is too soft. He means to say that people should be very careful with the answers. (And also, you don’t have to worry about the tools’ output if you’re not using the tools.)

This is a five-minute demonstration of a new feature in Adobe Illustrator that derives a 3D shape from a 2D vector. You kind of have to see it to believe it.

#ProjectTurntable | Adobe MAX Sneaks 2024 | Adobe by Adobe (YouTube)

Demonstrator Zhiqin Chen selected a vector, “generated views” for it (which took a few seconds), then was able to rotate it along both the horizontal and vertical axis to reveal that the tool had extrapolated a complete 3D shape from the vector. Wherever he left the shape, the tool continued to treat it as a 2D vector that the artist could continue to manipulate. Finally, he showed that, even after a shape had been cloned several times, manipulations of the original could be applied to the copies—in their respective orientations—by “updating the views”. All of the 2D vectors continue to be just that, no different from vectors that had been drawn manually rather than having been generated by the tool.

Turning a dragon in 3d spaceThe tool uses voodoo to “pull” a 2D vector up into 3-dimensional space, then lets you choose how to map it back into 2D space. The model remains in the background, allowing the user to continue to choose a different extrusion at will—until, presumably, the link to the 3D space is broken by changing the 2D view on it manually, in which case it becomes an untethered copy. From there, the user can generate a 3D view from the new 2D shape.

He showed how the tool was even able to derive four legs for a horse that had been drawn with only two legs. This suggests that the tool has a map to indicate to which part of the “3D-shape space” a particular 2D shape should be mapped. You already saw it with the dragon and the warrior, where the effect was subtler but essentially no different. The tool has to know that the large oval on the dragon’s belly should be belled out in 3D space. I really wonder how generally applicable this is, especially when using shapes for which the tool has less training material.

Still, the demo was very convincing—this seems like a powerful feature akin to being able to copy/paste text out of images and videos in MacOS and iOS.

There is an article in Microsoft’s documentation called How Microsoft names threat actors by diannegali & Dansimp (Microsoft). That sounds interesting. How does Microsoft determine and label threat actors?

“Microsoft shifted to a new naming taxonomy for threat actors aligned with the theme of weather. We intend to bring better clarity to customers and other security researchers with the nex taxonomy. We offer a more organized, articulate, and easy way to reference threat actors so that organizations can better prioritize and protect themselves and aid security researchers already confronted with an overwhelming amount of threat intelligence data.”

Where Microsoft is utterly unwilling to help you is when a threat actor comes from any country other than official enemies of the U.S., Israel, or NATO. The only threat actors for which they have a taxonomy are:

• China
• Iran
• Lebanon
• North Korea
• Russia
• South Korea
• Turkey
• Vietnam

As Sapir-Worf would say: since we don’t have a word for it, it doesn’t exist. That, or Microsoft just categorizes any threat from the NSA, CIA, or Mossad—just a few examples among myriad others—as being from Russia, North Korea, or Iran anyway. They probably have a special die that they role to pick a scapegoat.

So, yeah, it’s neat to see that otherwise-serious researchers kind of just pretend that two of the biggest hacking nations in the world—U.S.A. and Israel—just don’t exist in that sense. Microsoft is an international company. International customers should be pissed off that they prioritize sucking up to the Empire more than taking their job seriously in the name of customers who aren’t in the U.S.

Even U.S. customers would be interested in knowing when the CIA or NSA is putting trojans on their servers, but they’ll never hear it from Microsoft. I guess U.S. and Israeli trojans are just gentle, digital kisses—homeopathic balms that delicately lift your data from your data stores for your own good. They’re not really threats at all, in that sense, which is why they don’t exist in the threat-actor taxonomy. That’s just logic, baby.

The article The Cult of Mac by Cory Doctorow (Pluralistic) goes hard on anyone who uses Apple hardware.

“It’s Apple customers who lose access to apps that can’t be viably offered because the app tax makes them money-losing propositions. It’s Apple customers who lose out on the ability to get apps that Apple decides are unsuitable for inclusion in its App Store.”

It’s never even occurred to me to have this on my radar because I don’t use the App Store for anything but finding a very specific app, usually one that I’m forced to download. Do you want to invest a second to whip me up too, or are you just going to dismiss me as an Apple acolyte out of hand? I know that Apple’s app practices are abusive and monopolistic, but what’s the alternative to their hardware?

I’m caught in their hardware monopoly in that Windows is a dumpster fire and so is all of the noisy, energy-gobbling hardware that it runs on. iOS versus Android is the same. The hardware is quite significantly better. I’m all for putting pressure on Apple, but let’s not pretend that they have a stranglehold on the market just because they have an app-store monopoly. They actually make some pretty good hardware and decent services.

“These religious apologetics for Apple’s business practices are a devastatingly effective defense against the public outcry that would accrue to any other business that abused its customers in similar fashion. Every time Apple finds a new way to rip off its customers, the cult is there to insist that those aren’t true Apple customers at all!”

“[…] your old gadget gets “recycled” by Apple, who – uniquely among electronics manufacturers – drops all its “recycled” gadgets in giant shredders, ensuring that parts from old phones don’t find their way into the secondary market for use by independent repair:”

Apple iPhone 4sWhat an odd claim. I’ve never had a new iPhone. I’ve had four of them: an iPhone 4s and iPhone 5s, both hand-me-downs from my sister, an iPhone 6s bought from Revendo, and an iPhone 12 Mini, also from Revendo. Where did they come from if Apple shreds everything?

The 12 Mini is my current phone. It was introduced to the market 4 years ago. It works extremely well, even though it’s still on its first battery. I’ll almost certainly put a new battery in it within about a year and continue to use for several years.

“If it were the case that No True Apple Customer would patronize a third-party repair depot, then Apple could simply step out of the way of right to repair campaigns and those independent phone fixit places would sink without a trace.”

Some of them almost certainly would sink without a trace. Have you tried them? I had to leave one because it was so scammy. It would have cost three times as much as Apple and they wanted my password. Given that experience, you can’t ignore the downsides of opening up to competition: ads, scams, etc. I wouldn’t use the third-party stores, unless they had a really good reputation, because I’ve seen what that world does with people’s time and money. I have bought the last two laptops for my household (2 in ten years) from a third-party vendor, as well. I wonder if things are just different in the U.S.? (You know, in the land of the free?)

“Apple blocked Facebook from spying on you, but when it wanted to build its own surveillance advertising empire, it switched iOS spying back on, gathering exactly the same data as Facebook had, but for its own sole use, and then lied about it”

“One of the clinical signs that someone is in a cult is that they are encouraged to isolate themselves from people who aren’t also in that cult:”

Or it could just be the least shitty of shitty options. Again, I feel like Doctorow fails to see what the world is like outside of the California bubble. Internationally, SMS is a costly train wreck anyway, so the only alternative is to just get a different messenger if you want to communicate with the United States. There was never a useful alternative. If Apple were to make a perfect messenger, then he’d probably bitch that they’re using their monopoly power to squeeze independent messengers.

I like Signal. I would use it for everyone and drop Apple Messages, but some people are deep into the network effect. It’s hard enough keeping them from trying to contact me with Facebook Messenger or WhatsApp. Only Signal and Threema are quasi-independent of giant monopolies. And not nearly enough people are on that. And Threema’s desktop solution for messaging is an absolute dumpster fire.

“The company claimed that there was some nonspecific way in which Beeper Mini weakened the security of Apple customers, though they offered no evidence in support of that claim. Remember, the gold standard for security claims is proof-of-concept code, not hand-waving.”

The gold standard for proving that you are secure is not having software “based on a determined teenager’s code” FFS. Beeper was and is almost certainly leaky as shit. What makes you think Beeper’s code was secure? Literally no reason, other than, if Apple says it is, they must be lying. Everything is leaky as shit. The answer to Apple should be: then make a version that isn’t leaky as shit. Even they probably won’t be able to do it (they’re leaking your contact information via AirDrop right now).

As an update, Apple is rolling out RCS support for Apple Messages, even though I haven’t seen it yet. This is probably the alternative to which they were alluding.

To sum up: some of us are trying to navigate the corporate options available without conceding everything on functionality or hardware. Richard Stallman is absolutely admirable for having consistently done so but I am not that strong. Sometimes I just want a laptop that works for me rather than the other way around. I know he thinks his laptop works for him because it’s free but the reality is that the free solutions continue to require a lot of time in order to get to the point where you’re doing the task you originally wanted to do, rather than doing tasks required by the software and hardware choice you’ve made.
 

I was chatting with someone about some picture and I noted that maybe we could find something appropriate.

I prompted DuckDuckGo with “raccoon trash panda digging in garbage can”. It gave me a whole grid of pictures, of which I quickly picked the following as my favorites.

Trash panda relaxing in a garbage can as if were a hot tubTrash panda peeking out of garbage canTrash panda living its best life in a trash canTrash panda head down up to its hips in a trash can.jpg

My interlocutor had never really used any of the LLM-based machines before, so I gave GitHub Copilot a whirl. I prompted GitHub Copilot: “Show me a cartoon of a raccoon digging in a trash can with its butt in the air”. It gave me the following:

Raccoon in a flowerpot with a monkey's assRaccoon in a trash can with dangling feet

The second one is not bad at all, but the first one! What the heck? Is this what Copilot thinks I want when I ask for a cartoon of an animal? Maybe most people using these things are there to sexualize animals?

At any rate, I figured it should cover it up so I can use the graphic in SFW situations.

Prompt: “Put some orange swim shorts on him”

Raccoon in a trash can with an orange swimsuit and an offset tail

As I’ve experienced every other time I’ve tried generating images, there is just no way to get it to build on the previous image. You can’t include it as a template. A human artist would have given me the first two images, but with an orange swimsuit. Copilot gives me a completely different graphic of a raccoon towering over a garbage can, but with orange swim shorts, as requested.

Its tail is kind of just pasted on. Why? Because Copilot has no idea what a tail is. There’s just a high probability that there’s something stripy at the other end of a raccoon from its face.

The images I got just on the first page of DuckDuckGo are all way better. That took me seconds. Generating images takes a lot longer and never really gets to the desired result.

Apple keeps coming up with new things to mess with my typing. I long ago turned off auto-correct, but was surprised to see that my MacOS Sonoma Notes app started not only predicting text, but also auto-correcting it. I do not like this. I turned off auto-correct system-wide for a reason. I would rather correct my typos on my own. Just underline the errors and I’ll get to them. That’s my workflow.

These are my OS-level settings. I know I seem ungrateful to be turning off all of the assistance that Apple wants to give me, but if there’s one thing that I can do better without software sticking its grubby fingers in, it’s write.

Keyboard input settings

However, Notes—and I think Mail and maybe another Apple app or two—has another setting that turns on a special new kind of auto-correct that Apple really wants you to try.

Correct Spelling Automatically

No, thank you.

A recent experience at work led me to conclude that the AI revolution will pass most of us by. In mid-December, I fell ill with COVID. I’d updated my status in Microsoft Teams accordingly.

About six weeks later, a co-worker wrote to me, asking whether the status still applied? He hoped not?

I’d forgotten about it, but nothing had reminded me. It’s interesting that I get five mails a week about MS Viva and about Sharepoint Stuff I Might Have Missed, but I don’t get a single hint that my status might be out of date after six weeks.

So much for the AI revolution. This incident helped me refine my opinion on it. It’s definitely coming. but it will not be useful to me. If it is, it will only be incidentally or accidentally beneficial to me.

The prime use of AI will be of benefit to others, probably scammers or data scrapers.

You can set a custom time for the status-update options. But how am I supposed to know in advance how long I want to set it for? Sometimes you just don’t know. Wouldn’t it be nice if it would ask you after a day or two? It might sound annoying, but not if you include an option to “never ask me again for this status.” You could also just have an “ask me again when it seems stale” option or “how long do you think it should be set like this?” or “when would you like me to ask you about your status again?”

It wouldn’t even take AI to have a trigger that asks again after a week, unless you’ve told it otherwise. The likelihood that a status applies for that long is pretty low.

No, instead, Microsoft is measuring how long I spend in planned meetings and telling me how much “quiet time” I’ve had in the last month rather than helping me not look like an idiot who’s had COVID for two straight months.

Simon Willison continues to plug along, examining every LLM-related announcement and trying it out on his own machine wherever possible. The following video is a presentation he gave in early August. It’s quite interesting and worth the ~40 minutes.

'Catching up on the weird world of LLMs' − Simon Willison (North Bay Python 2023) by Simon Willison (YouTube)

At some point, he says:

“This is Vicuna 7b. It is a large language model. It is a 4.2GB file on my computer right now. […] If you open up that file, it’s just numbers. These things are giant, binary blobs of numbers—and anything you do with them just involves vast amounts of matrix multiplication. And that’s it. That’s the whole thing. It’s this opaque blob that can do all sorts of weird and interesting things.”

His description suggests mystery, but he’s really just described an executable file with machine code in it. Actually, he’s described any file, which, absent any form of inferred encoding that we consider to be “human-readable”, is just 1s and 0s.

It’s actually … what is it? A file is actually just a set of bunched electron configurations in a special material, where we interpret the bunched parts to be 1s and the sparse parts to be 0s. We interpret those 1s and 0s as a pattern that we call a “file system”. The structure is a language that we’ve invented to express complexity.

There are several layers of it. The material contains these bunches and we have circuits to read out these bunches reliably. Those sequences of 1s and 0s are interpreted bytes, interpreted through the lens of 2's complement, from which we derive numbers of various sizes. Some of those numbers we call characters, that we interpret with a specific encoding.

The only difference being that we understand the instruction set of the machine code, we understand the virtual machine for which it forms instructions. We ought to: we built it all.

The LLM, on the other hand, is an opaque runtime that we don’t really understand, in the sense that we didn’t design the circuits or the instruction set. All we know is that it has an input and output system onto which we can build plugins that allow us to use natural language to poke it, and to interpret its results as natural language.

It’s a mysterious process, but not for the reasons implied by the description above. A giant heap of numbers is a description of any file, even text files. The only reason we understand them as “text files” is that we assume an encoding for the 1s and 0s and derive meaning from there.

Notes:

• The analogy he draws between the iOS Keyboard prediction and LLMs is good. It’s just a matter of scale.
• It’s interesting to see how he uses the LLM in his daily work
• He also shows tools that he’s written that incorporate LLMs (e.g., one that reformulates his queries as more sophisticated prompts that are more likely to return usable results)

• He mentions several times that people are just poking around at these things, but there is little rhyme or reason to it. He cites one example of how it took two years for someone to discover that the model returns more reliable answers when you ask it to “go step by step”. There might be a plethora of other goodies like that hidden in there—or there might be nothing.

  I am, once again, reminded of Roadside Picnic.
   

• He goes on to discuss the data that contributed to it, and how he’s “very concerned” about the provenance of most of it.

  He doesn’t get into it more than that, but I will. Essentially, the same companies that will sue the ever-loving Christ out of anyone who uses anything of theirs that they claim to have copyrighted now simultaneously claim that their complete and utter disregard for copyright protection is obviously the thing that we want to do, because otherwise how would we even get all of this awesome stuff from which we’re hoping to profit immensely?

  So, they’re basically arguing that they can steal content from everyone without actually allowing anyone else to participate in this glorious world in which it’s OK to use each other’s content without permission. A nice trick, available only to very wealthy companies and individuals.
   

• He is one in a long line of people who is impressed by the way that these models can translate to French (or whatever) when he can’t actually read French—because a cursory glance at the translation shows that it’s not at all an accurate translation of the original, missing many idioms, etc.

I use the JetBrains Toolbox to manage my handful of JetBrains apps. On Windows, it has to keep track of ReSharper, Rider, DotMemory, DotTrace, and DotPeek. There are various settings to check automatically, to download automatically, to install automatically, etc.

I’ve never had a problem on either of my Macs (the Intel iMac or the M1 MacBook Pro). On Windows, however, the toolbox has twice now filled up my entire hard drive with exact replicas of an update package.

I almost always hibernate my laptop, but sometimes I leave it running overnight if I’m in the middle of a long-running task.

In the wee hours, JetBrains toolbox takes advantage of the idle processor to start updating itself, and updating itself, and updating itself, and updating itself…

…until it’s eventually filled the entire system drive. As you can see in the screenshot below, it’s using a huge amount of space for the “Temp” folder. I didn’t expand that folder, but there are dozens, if not hundreds of copies of the exact same folder.

Cache / Temp using almost all of the space

All in all, the Toolbox updater had gobbled up about 113GB of space before physical constraints stopped it. These weren’t just big installers. There were over 300,000 files to remove. It took almost five minutes to clean up this mess.

Deleting 113GB / almost 300,000 files

Over the last several months, I’ve been asked for advice on password managers. I am not a security researcher. I can only tell you what I do, and why. My experience and context are that I primarily use MacOS and iOS, as well as one Windows laptop. I was a LastPass user for a decade, but switched this year to ProtonPass.

• I’ve made my peace with cloud-storage for my passwords because I think the convenience outweighs the risk.
• Browser integration with a plugin is very convenient, for both retrieving and automatically updating or storing passwords.
• If you have a lot of Apple devices, then Apple will auto-fill the password for you in many, many places.
• On iOS, you can choose which password manager to use (e.g., LastPass or ProtonPass).

Evaluation

At one point, I put together a small evaluation matrix for a colleague, shown below.

As you can see, I don’t know that much about BitWarden, nor am I familiar with the UI/UX of Google or MS Authenticator as integrated into their respective browsers. What I do know is that any browser-specific solution—this applies to Firefox as well—is quite limiting, in that you generally can’t use the password managers in other browsers or contexts. An ideal solution is to have passwords auto-filled in whatever context you find yourself in.

LastPass

LastPass works well, as you can see above. I used it for about a decade. Why did I stop?

• I already have ProtonMail and ProtonVPN, so I got in early with ProtonPass and was pleasantly surprised by how much cleaner and smoother the UI was than that of LastPass. I was able to easily export and then import all of my passwords.
• Shaky implementation of “ignore this site” that would constantly ask me to store a site every time I created a link in Azure DevOps.
• A few very public breaches that, while I’m sure didn’t reveal any of my passwords, still ensures that my data blob is out there for everyone to see. Fingers crossed 🤞 that their encryption implementation is solid.

What’s still better about LastPass?

• LastPass does auto-fill usernames and passwords more reliably.
• It also picks up password-changes more reliably.
• It allows you to paste credit-card information all at once.

Those advantages come with a caveat, though: sometimes it pastes too much information, or information into the wrong boxes. The whole “paste and go” has become increasingly shaky, as UIs change and no longer reliably present patterns that password managers expect.

ProtonPass

ProtonPass does everything that LastPass does, but with a few UI and feature improvements.

• ProtonPass allows you to set a timeout and pin code to re-encrypt your in-memory data. The pin code is easier to enter than your master password—and also easier to guess—but it makes it more secure. Your data in-memory isn’t encrypted with your master password (as it is at rest and in transit), but it is at least not in plain-text in memory.
• It also integrates an email obfuscator that actually seems to work with modern sites. [1] This lets me avoid giving my real email to any old site.
• The UX is more stable and friendly. I can copy/paste an item, then quickly return to the dialog and it’s still open to the same information. LastPass would always reset, so you’d have to search for a site, copy the username, … then search for the site again to copy the password.

Adding and updating credentials

As noted above, ProtonPass is less aggressive about picking up changed passwords or filling in existing ones. I imagine that this is because they try less hard to detect every possible pattern that a site might use. I kind of agree that it’s better to make a good, stable UI that lets you easily copy/paste information than to try to keep up with the myriad ways that web sites are making authentication inconvenient for password managers.

For example, when I updated a password recently, I noticed that LastPass would pop up and ask me to store the password … but it was kind of annoying because it would ask too often. But I noticed that ProtonPass doesn’t ask nearly as often—though always more, with each update.

Even if it doesn’t offer to let you generate password, you can generate one manually.

Generate Passwords

When you add a password with ProtonPass, it hangs onto it for a while in the “generated passwords” area:

ProtonPass recommends reversing your workflow. Instead of clicking in the popup to generate a password, you should go to the “change password” page, then click in the plugin and generate a new password from there. It’s pretty easy, and it can auto-fill it from there. Here’s a little video:

Generate Password with ProtonPass

A layman’s thoughts about password-manager security

I’m going to repeat that, while I’m quite technically adept, I’m not a security expert. I read a lot about it. I haven’t caught wind of any problems with the implementation of either LastPass or ProtonPass. I wrote the following to a colleague about a month ago.

The primary thing for any of these providers is to keep the data encrypted at rest, in transit, and, as much as possible, in memory. Decrypting should only ever occur on a client device. That information should only remain decrypted long enough to use it—with an appropriate timeout.

Both their algorithms are solid and the number of cycles they require ensure that it would take more computing power than anyone has centuries to crack with brute-force. In this way, they’re the same as any other provider, like Apple, Google, 1-password, BitWarden, ProtonPass, Keepass, etc.

That’s the most important protection that any of these providers give you.

The algorithmic protection is, as noted above, only as good as your password. The algorithms together with a strong password, protect you from brute-force cracking. If you have a weak password, then attackers don’t need to use brute force because they can guess your password more quickly, without trying them all.

Another protection they give is to not allow access to your data to anyone but you. LastPass has failed to do this for a certain number of customers. It’s unknown whether any of the other providers have done the same—they might have kept the data safe, or … not yet been outed.

It’s known that some people’s password data has been expropriated. This shouldn’t matter because it’s just a pile of bits that is undecryptable without the password.

If you had a weak master password at the time that your data was known to have been stolen, then you’re at risk. You still have a chance to change all of your important passwords, though, before someone guesses your master password and gains access to the current passwords. If you’ve managed to change your passwords before they guess your old master password, then getting access to your old passwords gives them nothing.

Some providers (and security analysts) recommend keeping your ball of data locally, never uploading to the cloud. This is much less convenient and not something I’m personally willing to do. I would rather have a super-strong password that makes obtaining my encrypted data useless.

Layers of security:

• Algorithmic protection
• Strong password
• No access to data

Algorithmic protection is only as good the password. “No access to data” is a secondary layer. If it’s compromised (as was the case for some users at LastPass), then you’re left only with algorithmic protection + strong password. If you didn’t have a strong password, then you’re compromised, but you should have never been relying only on “security by obscurity” anyway (i.e., the fact that no-one had access to your data but you).

So, to sum up:

• I prefer a cloud-based password solution for the convenience, despite the risk that the ball of data could be stolen
• Your data can be stolen from any cloud-based provider (not just LastPass)
• Choose a strong master password to prevent brute-force attacks
• If you didn’t have a strong master password when your data was stolen, then:

  • Change your master password NOW to avoid the same problem in the future
  • Change as many of your passwords as you can to new ones
  • Hope that no-one got there before you did
• If you had a strong password, then you don’t need to worry too much because your data is still safe. Only one of the layers has been penetrated.

Andrej Karpathy

This is a pretty compact and interesting overview.

[1hr Talk] Intro to Large Language Models by Andrej Karpathy (YouTube)

At 46:00, Andrej discusses some of the available jailbreaks or “prompt escapes” that are still available, even with the latest LLM Agents. [1]

He shows how to reformulate a query for making napalm by asking the LLM Agent to tell it a story his grandmother used to tell him about making napalm. Or how to simply convert your query into the exact same text, but in Base64 encoding, in which case the LLM Agent gives the answer you were looking for, “escaping” its alignment/training/biases.

You can also avoid the training by using a non-English language because the focus has been on avoiding issues with English. All of these attempts to stop prompt escapes are just addressing symptoms, not the base problem. This is probably because they don’t understand how the black box of the LLM itself works, so all they can do is to massage the input in the hopes of getting what they consider to be more acceptable output, or to massage the output as well.

Sean Carroll

This is a great analysis of the state of LLMs and LLM agents by a physicist/philosopher who’s very good at communicating and thinking about hard problems.

Mindscape 258 | Solo: AI Thinks Different by Sean Carroll (YouTube)

He argues as well that there is a distinct difference in the underlying technology of the LLM/neural network and the agents with which we actually have contact—which are an LLM wrapped with many, many layers of bias and training and guardrails.

We should be aware of two things: (1) That there are guardrails that very clearly delineate the information that you’ll get out of such an agent and (2) that these LLMs don’t have an concept of the world, they have no context, they are just incredible word-associators.

He gives several interesting examples of his interactions, in which he demonstrates that the tools aren’t very useful—and are actively harmful to actually learning something—when approaching real-world problems, rather than the toy problems that you usually see demonstrated.

He asked the LLM agent about a hypothetical version of chess where the board was on a cylinder. Any human familiar with chess would quickly see that the kings are now right next to each other, and that the game would be over on the first move, as the kings start off in simultaneous checkmate.

The LLM Agent, however, droned on and on about what an interesting innovation this would be and just made up a whole bunch of shit that had no relation to the question, but was vaguely related to chess. The LLM Agent is a student who’s never paid attention in class and is trying to bullshit its way through the exam.

Every once in a while, the YouTube algorithm throws up a bit of flotsam from the shipwreck of content that I very much like—and that I would never have otherwise heard of. In this case, it’s a short video (4:43) of Stephen Fry reading a letter written by Nick Cave on the subject of LLMs and creativity.

Stephen Fry reads Nick Cave's stirring letter about ChatGPT and human creativity by Letters Live (YouTube)

I’ve citing at length below from the original blog post Iss #248 by Nick Cave (The Red Hand Files), which answered the question, “[…] what’s wrong with making things faster and easier?”

“ChatGPT rejects any notions of creative struggle, that our endeavours animate and nurture our lives giving them depth and meaning. It rejects that there is a collective, essential and unconscious human spirit underpinning our existence, connecting us all through our mutual striving.

“ChatGPT is fast-tracking the commodification of the human spirit by mechanising the imagination. It renders our participation in the act of creation as valueless and unnecessary. That ‘songwriter ‘you were talking to, Leon, who is using ChatGPT to write ‘his’ lyrics because it is ‘faster and easier ,’is participating in this erosion of the world’s soul and the spirit of humanity itself and, to put it politely, should fucking desist if he wants to continue calling himself a songwriter.

“ChatGPT’s intent is to eliminate the process of creation and its attendant challenges, viewing it as nothing more than a time-wasting inconvenience that stands in the way of the commodity itself. Why strive?, it contends. Why bother with the artistic process and its accompanying trials?

“[…] even though the creative act requires considerable effort, in the end you will be contributing to the vast network of love that supports human existence. There are all sorts of temptations in this world that will eat away at your creative spirit, but none more fiendish than that boundless machine of artistic demoralisation, ChatGPT.

“[…] It is our striving that becomes the very essence of meaning. This impulse – the creative dance – that is now being so cynically undermined, must be defended at all costs, and just as we would fight any existential evil, we should fight it tooth and nail, for we are fighting for the very soul of the world.”

Another post from January Issue #218 by Nick Cave (The Red Hand Files) first addressed LLMs, in what would eventually become the tour de force above, but which also had some wonderfully written prose about the difference between human creations versus those produced by imitation machines.

“What ChatGPT is, in this instance, is replication as travesty. ChatGPT may be able to write a speech or an essay or a sermon or an obituary but it cannot create a genuine song. It could perhaps in time create a song that is, on the surface, indistinguishable from an original, but it will always be a replication, a kind of burlesque.

“Songs arise out of suffering, by which I mean they are predicated upon the complex, internal human struggle of creation and, well, as far as I know, algorithms don’t feel. Data doesn’t suffer. ChatGPT has no inner being, it has been nowhere, it has endured nothing, it has not had the audacity to reach beyond its limitations, and hence it doesn’t have the capacity for a shared transcendent experience, as it has no limitations from which to transcend. ChatGPT’s melancholy role is that it is destined to imitate and can never have an authentic human experience, no matter how devalued and inconsequential the human experience may in time become.”

I unfortunately and occasionally end up on official support-forum pages.

You know the ones.

The ones where a community member or MS expert or Apple expert will tell you to restart your computer in safe mode because you asked why an app keeps losing focus when it shouldn’t. They will think of literally anything to waste your time, your life, but they will never cop to the actual problem you’re reporting.

Most of these answers don’t really relate to the question at all. It’s just a way for the user answering the question to gain points for having answered and for Apple or MS to be able to close the question for “no activity” after a little while.

The reason there is no activity is because what’s the point of engaging an answer that’s come from an obvious bot or human idiot who’s actively uninterested in helping you? In the end, none of these forums are there for the purpose of helping people—interested just there for companies to pat themselves on the back for fulfilling KPIs.

The answers are almost always along the lines of: have you tried it with another computer? With another user? Have you tried rebooting in safe mode? Have you tried uninstalling and reinstalling the software? How about your operating system?

THE UI HAS A FUCKING TYPO, YOU DOLT.

THE UI SCROLLS RANDOMLY, YOU UTTER WASTE OF OXYGEN.

I am not here because I think this bug is my fault.

I did not describe the issue in exquisite detail because I wanted someone to tell me how to unplug and re-plug my computer.

This rant was triggered by the following example, Apple Music automatically scrolls down playlist by Aleks_Bromvig (Apple Community). This poor person—the one who originally reported the issue—succinctly explained what was happening. I was one of about 30 people who clicked the “Me too” button to indicate that they’d described the problem that we were also having.

“When I rate music or add a comment the program will suddenly scroll all the way to the bottom of the playlist. I have been having this issue for quite some time and I can see other people asking about this problem seemingly without having their issue resolved.”

They’re right: this has been happening for version after version of Apple Music. They’re absolutely right. What was the response? Well, the indubitably human “Community Specialist” Grace2211 responds and I’m going to quote in full,

“Thanks for reaching out in Apple Support Communities. We understand that the Music app is not working as expected when trying to enjoy your Apple Music membership.

“To clarify, do you experience this behavior with another device as well?

“We’d like to provide you with a few steps that may help resolve the issue and allow us to determine if the issue could be software related:

“• Verify if the issue persists in safe mode. Your Mac will load slowly and your screen will flash. This is expected behavior for this mode:
How to use safe mode on your Mac

“• Test the issue in another user account: Set up users, guests, and groups on Mac”

If this is just a friendly ESL person helping out, then fine. But my experience has been that this is the only kind of response you can expect. It’s a poorly written, overly wordy nothingburger of a response that blames the user for the entire problem, when it’s an obvious bug that has been happening forever.

It covers all of the bases: it starts with a useless, time-wasting paragraph: “We understand that the Music app is not working as expected when trying to enjoy your Apple Music membership.” There is nothing more infuriating than this. It just means you didn’t read the question. You’ve copy/pasted this bullshit from a standard template. I wasn’t trying to “enjoy my membership”; I was trying to use your buggy product, for which I pay money every month.

This is only going to get worse as companies ramp up their production of such useless copy with LLM-produced text that desperate users will be forced to wade through in the dim hope that an answer for their problem lies buried in the tsunami of meaningly and soulless text.

But I digress.

Then, it asks you to try another device—buy another Apple laptop to see if Apple Music is just as shitty on that one!—then says “hey, what about a long, involved reboot into a crippled-Mac mode in which Apple Music probably doesn’t even work?”—waste tons of your own personal time rather than a second of Apple’s support time—and ends up by telling you that, if you only have the one device, you should set up a completely new user account, log in with all of your Apple stuff so that you can even access your music and … then what?

Of course and obviously, Apple Music is going to do that stupid scrolling shit there, too, because it’s obviously intrinsic to the software.. It has been for years.

The only answer you should give is:

1. Ask the user if any other actions cause the list to scroll, like maybe “liking” a song? Does that cause the scrolling? Or just the by-now frowned-upon 5-star rating system that has been in Apple Music forever, but which is only available in the MacOS client in anything approaching a usable manner? [1]
2. Tell the user about a plausible workaround, if there is one.
3. Tell the user that you’ve sent the bug report to the relevant team within the organization and are awaiting feedback.

But my experience is that, the bigger the company, the less likely it is to have an in any way responsive support for its software products.

By the time I got to the page, it had already been helpfully marked with something like,

Thanks, asshole. I’d love to start this conversation anew, so that you can continue to make it look like no-one else has ever had this problem.

The only thing useful on that page is the “similar questions”, of which one was actually similar enough to be worth clicking on, so there’s that. That took me to Mac Music app bug?, where the helpful user “turingtest2” pointed me to Songs View Jumping to Random Spots After Action instead of trolling me to buy another computer, restart in safe mode, or set up another user.

That page is from over two years ago, has an astonishing ~850 “me too”s on it—and the bug still hasn’t been squashed by Apple’s veritable army of engineers, at least in the version that everyone’s asking about. It actually has been squashed on the newest version of Apple Music. 🍾🍾🍾 I reported this in that forum, to lend hope to the ~850 other users who are looking for an answer.

“The title says it all. When I rate a song with either the stars or the heart, the playlist jumps to a different scroll position in the list. This is happening in Apple Music 1.2.5.7 on MacOS 12.7.2.

“I can verify that it’s not happening on Apple Music 1.4.1.29 on MacOS 14.1.2, so you must have found and fixed it. Any chance of backporting it for those of us with hardware that works perfectly well, but can’t be upgraded to the latest Apple Music (which is bundled with an OS not supported on this hardware)?”

Just in case you think I’m raging about an isolated instance, here’s another example, encountered just in the time since I created this draft. This one is about another perennial favorite: Notes app lagging by Brittlopez, which writes:

“Is anybody else experiencing lag when typing in notes. I’ve closed all apps prior to using it I’ve tried resetting and I’ve used a different apple device. Typing everywhere else is fine but typing in the notes is so slow.”

Look, this has been happening for years. The goddamned app sometimes seems to sync about forty times per letter typed. I’ve had it drain most of my phone battery within a half an hour, simply because I was writing a journal entry—because I didn’t have my laptop handy and I was inspired.

It shouldn’t be this hard to write a collaborative tool, but it apparently is. To add insult to injury, if you type a bunch and it feels like it’s sucking your battery dry just to sync a few paragraphs, then that text still doesn’t show up on other devices. You have to start typing in that note on the unsynced device, then wait several seconds. The synced text will finally pop in. It’s quite sad, actually. But, sure, let’s make Freeform and Journal because if you can barely sync text over a gigabit connection, then you should be ready for all sorts of even meatier content.

But I digress. Again.

Anyway, the people on these forums are just castigating themselves for having a few photos and some longer text. Stop it. You are not the problem. The problem is a $3T company that can’t seem to get its shit together to make an app that syncs reasonably well. There are real-time collaboration tools out there, and Notes is still a dumpster fire. I use it because it works for me and I don’t sync that much—it’s not a main tool, by any stretch of the imagination.

Anyway (again), the top response in the forum is as follows:

Reboot instructions

“Try and Force Restart your iPhone EXACTLY as shown below and see whether that resolves the issue:”

• Press and quickly release Volume UP button
• Press and quickly release Volume DOWN button
• Press and Hold the SIDE button until an Apple logo appears and then release the Side button (Can take up to 20 seconds. (DO NOT release Sid

The comment is cut off, but you get the gist: they just copy/pasted the instructions for restarting your iPhone. That’s the solution? Do I also have to restart my Mac when Notes gets laggy? That’s the only option? Am I to reboot my devices as a Hail Mary every time one of your buggy apps misbehaves? Then, by the time I’ve rebooted, I’ll have forgotten why I did so? Is that the hope? Can’t you just fucking fix the bug?

The tl;dr is that the current desktop client has been in maintenance mode for almost four years. It requires that an iOS phone be available, connected, unlocked, that Threema be in the foreground, and that the screen be on in order for the desktop client to function at all. [1]

The successor—Threema Desktop 2.0—has been in development for about 4 years, has been actually available in beta for about a year, and is still so buggy and limited in functionality as to be barely usable. The bar set by Apple Messages and Signal seems be far too high for Threema.

Even the iOS client still doesn’t have inline previews for links—how hard is that? Is it somehow insecure to download a picture and show it?

Desktop client is high-priority for me

The iOS client is otherwise fine, but I would really, really, really like to use a keyboard for text and drag-&-drop for links and pictures. Especially since I’m sitting at a computer most of the day anyway.

With Signal and Apple Messages, I can respond comfortably, whereas with Threema, I have to pick up my phone because the desktop client is either too tiresome or isn’t available because I can only link one of my desktops/laptops.

Recent history of the desktop client

The web-based desktop client is an absolute phone-killer on iOS now. The app has to be open and the phone has to be unlocked and the screen has to be on. This client was placed in maintenance mode on June 15, 2020, about 3.5 years ago. The readme states,

“⚠️ Note: Threema Web is in maintenance mode while we are working on Threema for Desktop 2.0, which should resolve some of the long-standing issues we were having with Threema Web. We will still do regular maintenance and fix critical bugs, but for now there will be no major new features or non-critical bugfixes. See #996 for more details.”

The link to #996 takes you to a sad list of comments June 2020 through October 2021, with basically no progress. In late January of 2022, I learned from Threema during a job interview [2] that Desktop 2.0 was on the way. 🎉 🙌🏼

📆 📆 📆…

I commented on July 1st, 2022, asking about the status. I included the following screenshot I’d found on Twitter.

threema desktop 2.0 screenshot

They wrote back nearly immediately to say they were “working on it like crazy.”

I responded with,

“Thanks for the quick reply! “Working on it like crazy” is enough of a sign of life for me, TBH. I just noticed that it had been six months since I’d heard anything (screenshot/tweet above) and that the initial announcement of the new architecture had been made 2.5 years ago.

“As for my experience with the current desktop, I regularly see the open desktop app, with a conversation selected. I can even write a message, but when I submit, it shows the “Reload Session” button and throws away the comment. I’ve gotten used to just clicking a different conversation first to see whether I’m still “logged in”. That way I don’t lose any message text. It’s just not ideal. Signal is much more comfortable as a desktop client (although their nearly daily updates are also not super-reassuring). I have an old iPhone 6S (although I doubt that plays into it).

“Maybe I’m the weirdo for using a desktop/web version, but it’s so much more comfortable to use a real keyboard and to be able to copy/paste from a real browser. I know a lot of people are exclusively on their phone, though.

“Hey, I’m happy to hear you’re working on it. I also understand if you’re going with the old Id Software roadmap: “when it’s done.”

“Thanks again for the quick response.”

The response kinda/sorta corrected my note that it’s been 2.5 years by writing “1 ½ years since the announcement. But obviously there was internal work before, so your number is actually close. 🙂”, confirmed that the “[t]he iPhone experience of Threema Web is indeed rough,” and that “the whole Threema team would like to have released Desktop 2.0 already”.

There was also a relatively honest block of reasons for the delay.

“Besides technical challenges on the Desktop 2.0 code base itself, let me name a few other reasons why it takes so long: Multi-Device protocols are very complex if one wants to prevent desyncs and conflicts. Another challenge is to update the existing code of the apps so that they harmonize with each other. This is vastly more difficult than writing another Threema client as they exist today. And I think it’s fair to admit that, even though we were prepared, the magnitude of it still took us by surprise.”

All understandable. That was almost 1½ years ago. The Threema Multi-Device: An Architectural Overview white paper was written in November 2020 and doesn’t seem to have changed since then.

Working with the preview

At the beginning of this year—6 months later—a preview version became available, which you could test if you upgraded to a beta version of the Threema client on iOS. It was limited, but worked OK for a while, then stopped working after a few months because things got a little out-of-sync. I stopped using it, then decided to reinstall everything again and got it working.

A day or two later, I had strange issues with messages not appearing on iOS or the Desktop, and some friends said they weren’t seeing my messages. On top of that, the iOS client started constantly spamming dialog boxes to tell me that there was another device linked and the client would no longer update. Yes, I know. I linked it on purpose. You not only watched me do it, but you helped me do it.

As soon as I unlinked the Desktop device, everything went back to normal.

I still have friends on Threema only, so I recently reinstalled the desktop client again. There is still a list of limitations a mile long.

• The beta version of Threema 2.0 for desktop only works with one iOS device and one computer (Windows/Mac/Linux).
• It’s available for iOS only (Threema for Android is not yet supported).
• […] videos are sent as file messages. Recording voice messages and sending or receiving polls and locations will be supported in a future release.
• New message notifications may appear on your iOS device even if you have already read the messages in the desktop app.
• The “Typing Indicator” and “Do Not Disturb” settings are not yet supported by the desktop app.
• Perfect Forward Secrecy (PFS) is not supported when using the beta version of Threema 2.0 for desktop.

There’s also no support for audio or video calls, obviously.

At least one friend noted that PFS had been disabled and whether he should be worried? I had to admit that it was because I’m using a beta client from Threema and that they turned it off because they don’t seem to be competent enough to get it working in multi-device mode within a reasonable time span.

It’s incredible. After at least 3 years of publicly known development work (and 4 years of admitted work by the developers), it barely works. It’s stable enough, but it’s still very much an alpha, in that it can’t possibly be considered feature-complete.

This is what the client looks like today, in December of 2023.

Threema Desktop 2.0 Beta in December 2023

Note that it looks nothing like the screenshot of the purported functional prototype from December 2021 (shown above).

I’m kind of embarrassed for them.

Conclusion

Signal has a desktop client. They’ve had it for years. I regret ever having gotten anyone on Threema because the desktop solution is so MIA right now. Signal does everything I need and it’s at least as secure—if not more secure—than Threema. Now that I see how shakily they’re developing their desktop software, I’m losing confidence that they’re any better at the security side.

It just feels like Threema isn’t really good at developing software. Maybe they have good developers, but they don’t know how to ship software [3]—or they absolutely don’t care about this particular piece of software.

Maybe they’re focused on their corporate customers, who actually pay them for their product. That’s fair. But the least they could do is admit that the multi-device desktop client for the rest of us just isn’t happening.

With LLM-produced materials, we are currently forced to rely on belief that what we ask for is what we will get. We don’t know. We can’t prove it.

For example, image generators have been given billions and billions of images and pictures of people and still they generate material with people that have three arms and eight fingers. There are guardrails in place in most image generators, but the LLM at the core of the machine doesn’t know anything. It doesn’t know that people don’t have three arms.

A child learns very quickly how many fingers and arms a person has. An AI does not. It has to be explicitly taught these things. This is fundamentally different. You put a human in the world for a while and it learns a tremendous number of things that we’ve taken for granted. An AI does not do this. You have to hold its “hand”.

This is OK with stuff that we can easily verify. But what happens when it’s not easily verified? What happens when it’s an MRI output or something much more complex and difficult to verify? Then we have to take it on faith that each of those shadows on the MRI is really a potential tumor and not just three extra fingers.

I just don’t think this technology is a viable path forward. I think that this technology base doesn’t scale. We’ve managed to scale it pretty far with bloody-mindedness, but we can’t seem to refine it. Well, we can refine it, but we have to teach these tools every single little detail about the world so that we can better believe their output.

But just that we have to “believe” as part of our use of a technology is very, very different than scientific approaches. It’s fundamentally different than other tools we’ve built. In other cases, most people had no idea how their tools worked, but at least someone did. It was possible to learn. With LLMs, the black box is a black box for everyone.

On top of that, an emergent feature of these things is hallucinations—like the eight fingers and three arms—but also a tremendous inherent bias in the input, for which there is no reasonable solution. Because how do you correct a bias? To what would you correct it?

If I ask an image generator for a a hot slut, it will 100% produce a nearly-naked woman, not a man. It will also not produce androgynous content. It dreams, but it dreams the dreams we taught it to dream.

As noted in the tweet On the hallucination “problem” by Andrej Karpathy (Twitter),

“[LLMs] are dream machines.

“We direct their dreams with prompts. The prompts start the dream, and based on the LLM’s hazy recollection of its training documents, most of the time the result goes someplace useful.

“It’s only when the dreams go into deemed factually incorrect territory that we label it a “hallucination”. It looks like a bug, but it’s just the LLM doing what it always does.

“[…] An LLM is 100% dreaming and has the hallucination problem. A search engine is 0% dreaming and has the creativity problem.

“[…] An LLM Assistant is a lot more complex system than just the LLM itself, even if one is at the heart of it.

“[…] the LLM has no “hallucination problem”. Hallucination is not a bug, it is LLM’s greatest feature. The LLM Assistant has a hallucination problem, and we should fix it.”

Everything it does is hallucination. Some of it happens to hit close to what we consider to be a bullseye.

The discussion in the video below expands on that point.

Making Large Language Models Uncool Again by Outerbounds (YouTube)

The following transcription is from about 25:30:

“So something everybody I think pretty much agrees on, including Sam Altman, including Yann LeCun, is LLMs aren’t going to make it. The current LLMs are not a path to AGI. They’re getting more and more expensive, they’re getting more and more slow, and the more we use them, the more we realize their limitations.

“We’re also getting better at taking advantage of them, and they’re super cool and helpful, but they appear to be behaving as extremely flexible, fuzzy, compressed search engines, which when you have enough data that’s kind of compressed into the weights, turns out to be an amazingly powerful operation to have at your disposal.

[…] And the thing you can really see missing here is this planning piece, right? So if you try to get an LLM to solve fairly simple graph coloring problems or fairly simple stacking problems, things that require backtracking and trying things and stuff, unless it’s something pretty similar in its training, they just fail terribly.”

We’ll have to reevaluate the tech and try again. I imagine we’ll futz around for a while first, letting some fools get spectacularly rich on it first, as we always do.

Back at the end of August, I read the article Making Large Language Models work for you by Simon Willison. I have since being doing much more research about integrating LLM-based assistants into the development workflow for work. It’s quite interesting, and I’m going through some older content to see what’s worth mining for that effort.

In particular, the article has this description of expertise, and linked it to ChatGPT—obviously, it’s Simon Willison.

“LLMs have started to make me redefine what I consider to be expertise.

“I’ve been using Git for 15 years, but I couldn’t tell you what most of the options in Git do.

“I always felt like that meant I was just a Git user, but nowhere near being a Git expert.

“Now I use sophisticated Git options all the time, because ChatGPT knows them and I can prompt it to tell me what to do.

“Knowing every option of these tools off-by-heart isn’t expertise, that’s trivia—that helps you compete in a bar quiz.

“Expertise is understanding what they do, what they can do and what kind of questions you should ask to unlock those features.”

Well, welcome to the party. Expertise has always been exactly what he’s described. It’s having an understanding of a subject—wisdom about it, if you like—born of extensive familiarity. But it’s never been about rote memorization of things. How is this a revelation?

Sure, experts tend to have to look things up less, just because they’ve done something you’re asking about so many times before that they can’t help but remember how it’s done. My expertise in programming techniques, programming languages, and development environments leads me to expect more, to be able to conceive of a feature I’d like to have and to go looking for it. A lot of people can’t do that because they don’t know what to look for. Therefore, they’re not experts.

The only thing that really is about deep familiarity and rote memorization is vocabulary, the toolbox from which you draw in order to express your thoughts. When I want to type a word like “morass” and I can’t remember whether it has two r’s or two s’s—or both—and then use a real-time spellchecker to test which version is correct, only to realize that it doesn’t have an ‘e’ at the end, I’m still expressing my own thoughts, in words that I know. If I don’t know the word, I’m going to express myself more simply—albeit perhaps more understandably!

When I use an LLM to generate entire swaths of text, I’m no longer expressing anything of myself. It’s not my thoughts. It’s words generated from a kernel that came from me. It’s leveraging, sure, but it’s a fundamentally different expression. It contributes much more text—which others have to wade through—from much less, not only effort, but much less thought. You’re essentially cheating people, tricking them into reading what you’ve gotten the LLM to write for you.

So, yes, expertise ineluctably comprises at least one skill: an expert is someone who’s amassed a formidable arsenal of tools with which to express their thoughts. If you don’t have thoughts, you’re not an expert. If you rely on tools to express your thoughts for you, then you’re faking it. However, you might be able to eventually fake it well enough to provide value to society? I don’t know if that’s true, but I’m willing to give it the benefit of the doubt.

There are some tasks for which immediately available, immanent expertise is essential, where the ability to quickly correlate information from disparate sources is exactly what the interlocutor—or situation—is looking for. There are others where a delay is OK. Say, you need to know how to light a campfire. It’s great if you have someone in the group who already knows how to do that, but, you can also just look it up and learn how to build a fire in five minutes. If you need to know the temperature, likewise.

Where immanent expertise is important is when you don’t have a data connection. If your keeping your expertise off-site, then you run the risk of being cut off from it. You’ll all freeze to death if you’re each having a conversation over a spotty data connection with an LLM about how to build a fire, and it’s veered off into telling you about a really cool story by Jack London. It’s relevant to your imminent death, but irrelevant to your wanting to live.

[…] so under-equipped to be communicating in the first place that it’s a crap-shoot as to whether they can express or understand any concepts worth discussing

A task for which immanent expertise is currently very advantageous, if not essential, is debating, participating in meetings, talking to other people. The thing that greases the wheels of civilization, in other words. Being able to properly express what you’re thinking in real-time is helpful. The current idea of offloading to a web search or LLM prompt incurs too much delay to be a viable replacement, or even an alternative.

Can you imagine it? Instead of learning a language, with vocabulary and practice in elocution, one party expresses a truncated set of half-baked bullet points that they balloon with an LLM into several paragraphs of text that they then send, unread, to their counterpart, who sends the text, unread, to their own LLM, which distills it back down to a few bullet points, which, one hopes, bear some semblance to the original ones, but it doesn’t really matter because both parties are, at this point, so under-equipped to be communicating in the first place that it’s a crap-shoot as to whether they can express or understand any concepts worth discussing.

All that said, and I honestly can’t see the advantage of having an LLM answer these questions rather than a search engine. I manage to quickly extract answers from DuckDuckGo every damned day without feeling like I’m restricted because I didn’t get to ask an LLM 12 questions to refine the answer, or ask the search engine to answer as a goat in a tree (something that Willison above thought was a fun thing to try). What absolute madness is this?

What’s mind-boggling is that this is a very smart guy who only hit upon the idea to use a tool to “remember” Git commands for him when he could do it with an LLM. He still uses Git from the command line, but he now pipes his questions through an LLM first—e.g., he asks it how to “undo last Git commit” and it tells him git reset HEAD-1 (which, honestly, seems kind of intuitive enough to remember)—and then executes it on the command line. And then he calls this “efficient”. I’m blown away that he’s never heard of a Git UI. I just type Ctrl + Shift + K from long years of muscle memory using SmartGit.

This is a question I have for anyone who asks me about how to leverage LLMs in programming: are you even using the other tools we already have available? Are you just grabbing for the new shiny? You know, which you will use just as spottily and badly as all of the other tools that are already at your fingertips and which you ignore daily?

I’ve watched Mo Gawdat before (see Mo Gawdat discusses AI). He is an acquired taste, at least for me. There is good in what he says, but it is interspersed with a lot of wild and unsubstantiated statements that he hopes you’ll believe because he’s so smart. The listener is left wondering whether they don’t see the through-line on what he’s saying because he’s skipped a bunch of steps that his unparalleled genius didn’t see as necessary or wether he’s just pulling a fast one. [1]

This is the video:

URGENT: Ex-Google CBO says AI is now IMPOSSIBLE to stop with Mo Gawdat by James Laughlin (YouTube)

And these are my cleaned-up, more-or-less, stream-of-consciousness notes I took as I listened to this video.

The interview starts off with a warning by the clearly overexcited host that the topics that will be discussed are so transgressive that you might be triggered by them. OK, sure. Whatever.

Then, there is the by-now familiar Mo Gawdat introduction where he talks about writing an entire book in nine days because his mind is so organized and his CHI is SO FLOW and he uses silence as a fucking weapon and he doesn’t waste time being like those other high-powered billionaire executives who are always chasing the cheese in the maze…but then he says things like,

“One of my best, best friends is Gelong Thubten, who’s one of the top monks of the UK.”

What in the hell does that even mean? Is there a FIFA-style ranking for monks?

I wouldn’t mention it but, for me, it reveals that his mindset isn’t quite where he’s like to have it yet, I think. But hey, no problem, life is aspirational. Talk about the thing you want to be until you are that thing.

Because what he is advocating is, in general, pretty good. But it also seems like it best applies to those who no longer have to worry about any worldly needs, those who have achieved financial orbit. Because not following that advice is what made him a hyper-millionaire in the first place.

For those who aren’t in that enlightened post-capitalist place—i.e., in the way that he’s used capitalism to escape capitalism—the advice may ring a bit hollow. Also, the dude is wicket smart, and it’s often the case that smart people can’t quite see why other people don’t just try harder to be as smart as them.

The host James is really embarrassing himself. He’s all like, “aw man, I would love to be silent for days,” to which Gawdat says, “even 26 days is not enough.” 🤦‍♂️

Cool, bro…so the podcast host wants to be silent more, and the orbital capitalist millionaire tells him that he should do more than 26 days of silence. Neat. Did Gawdat forgot that the system is organized in a fashion that most people can’t take that much time off without getting hungry or cold? Or that the guy he’s talking to is literally full of shit because his whole jam is to talk on videos for likes to make money?

Gawdat continues,

“By day 32, clarity sets in.”

Sure, ok. 32 days without “reading, inputting information, or interacting with people.” is … a lot. I feel like it’s the kind of thing that people do who can’t find balance otherwise, who can’t figure out how to get silent moments integrated into their normal lives. He talks about sitting in front of a paper notebook without any digital input, etc. But it would kill me to sit that long. Instead, I would go for a walk or a hike.

He does mention that he sometimes does “mini retreats” where he starts his day at 16:00. Sure, again, good for you. I don’t think the 7–11 employee gets that many mental-health days.

He talks about AIs “being smarter” than us and that AIs will be “a billion times smarter” than us “by 2037”. What the hell does that even mean? I like that he doesn’t even consider that he might be wrong about these levels of smartness. Like, where does context and wisdom enter into it? Like, what about useful intelligence? If you’re capable of grasping incredible complexity, but you don’t know a language that anyone else knows, then it’s of limited use.

I find these discussions interesting, but I don’t know what that has to do with LLMs. It can get a PhD, it “outsmarts us”, but it still doesn’t know how many arms a person has. It can be convinced that 2 + 2 = 5. Don’t we have to understand what this kind of “smart” actually means?

In a way, there are already such beings in the world. They walk among us. They are the smart people. Most people don’t grasp a goddamned thing about their world. Those who do grasp a lot—who are currently at the top of the heap—are they now terrified of being left behind? Of being like everyone else? Are we simply witnessing the panic of a self-selected intellectual elite being terrified that they’ve made themselves obsolete?

Are they scared of things existing that they don’t understand and can’t understand? That’s OK, no? There’s a ton of stuff happening in countries where I don’t know the language or the culture or anything. That’s all out of my control already. There’s no way I’ll ever understand it. I wonder how much of what Mo’s talking about is the terror of a control-freak?

The attitude he has toward AI feels, to me, conceptually similar to the attitude that the U.S. has to anything it doesn’t understand. Subjugate or eliminate. Maybe that’s the right attitude to have for AI as well. It might be the right one because this time it’s different—but, man, have I heard that story many times before. I suppose if you accept that premise of smartness—he still hasn’t defined it more than vaguely—then you’d want to keep it from replacing us? Are we really talking about that?

I think his comments in the other video were pithier—that it’s not the ASIs we should be afraid of, it’s what people will do to us with them. I fall back on my comparison to the development of atomic power plants…and then atomic weapons.

At 26:30, he says,

“one of the best code developers on Earth today is AI. As a matter of fact, with weeks or months or years—it doesn’t matter the time; it’s inevitable, it doesn’t matter when—they will be, by far, the best software developers on the planet.”

It kind of does matter when, no? Seriously, this guy elides so much stuff from his arguments. I wonder if he’s thought it through and he just skips large portions or whether he’s just … full of shit.

He just hand-waves away the temporal component. It doesn’t matter when? Like, if they became better developers millennia from now, that would be the same so-called threat as if they were already the best software developers? C’mon, dude.

He then cites another friend of his, the CEO of Stability.AI, who says that,

“40% of all code on GitHub today is written by a machine.”

First of all … proof?

Second of all … are we just going to take a CEO of an AI company at their word that AI is taking over?

Third of all, is Gawdat being sneaky when he says “machine”? There’s already a ton of generated code, but it wasn’t generated by an LLM. It was generated by tools that create boilerplate.

And if it’s 40%, is that good code? Or is volume the most important thing?

We’ve spent decades trying to escape the Charybdis of LOC and here we’re pulled right back to measuring by size, not quality.

I just want to note that James is insufferable. He offers no pushback at all. Nothing.

Gawdat again:

“10 out of 10 of the most beautiful women in the world are not human. They’re generated.”

C’mon, dude. You start off with this woo-ey meditation shit, but you think that a statement like that isn’t philosophically fraught? Isn’t beauty in the eye of the beholder? That people think an AI-generated person is beautiful … doesn’t that say more about the superficiality of our society than about a takeover of AI? There are so many better things to discuss than this angle.

“you have GPT being that you know geek boy nerd if you want or—and I say boy, sadly, not girl okay? Because, again, it’s developed around IQ and there is a lot of emphasis on the masculine side of analytical thinking and so on and so forth, which is an unbalanced form of intelligence.”

There’s a lot to unpack there. Analytical thinking is masculine? Well, well, well. This kind of attitude is, I suppose, the kind of thing that leads to the inherent bias of the machine that he’s talking about, but I’m increasingly less likely to give him the benefit of the doubt that that’s what he was trying to imply.

I find it interesting that people like Gawdat discuss humans and people and what they would do, all without really speaking about how they actually tick. He says,

“ I think when AI reaches that level of intelligence will become irrelevant to it. […] No human wakes up in the morning and goes ‘you know what? I’m so annoyed by ants I’m gonna kill every ant on the planet.‘ Nobody does that, okay?

“It’s just [that] ants become irrelevant. They become relevant if they come into your space, so you may spray your balcony or whatever but no human comes up with that enormous plan of ‘you know what? The world is bad until we get rid of all ants.’ Nobody does that.”

Ok. like, you’re ignoring a lot of history. People very definitely do that. It’s called genocide. They don’t always get every last one, but it’s shocking to hear someone so admiring of their own intelligence not even think about Hitler or Suharto or Armenia or Native Americans.

I wonder why he’s so laser-like focused on potential problems while ignoring all of the very real ones that we have now. Like, he’s worried about how we’re going to interact with an AI that will be all-powerful and indifferent to us, right? But there are billions of people on the planet who already live exactly like that. Their lives are entirely influenced and completely controlled by the whims of an unseen and unknowable elite.

It’s hard not to see Gawdat’s panic as being the reaction of someone who is in that elite and realizes that he may soon not be at the top of the heap anymore—as another alpha predator comes to town. Instead of recognizing the situation and trying to remedy his own role in it, he imagines a new layer and sounds the klaxon.

AIs are going to destroy us all. Um, yeah, I guess, those of us that weren’t already destroyed by capitalism? Like, capitalism’s utter inability to do anything positive about climate change. Austerity. Intensifying animosity and dis-empathy between peoples. And I’m supposed to worry about SkyNet?

I honestly feel like I’m listening to a blockchain huckster. The style is the same.

At 31:30, he starts talking about how “the most valuable asset on the planet … intelligence.” I was just talking about this conceit with a guy I met in a bar yesterday (Matuš). The problem is that our society values the wrong things. The most intelligent people also consider themselves to be the most valuable. Yes, intelligence can be leveraged, but everyone is important. That intelligent person doesn’t help anyone if they die of sepsis. They’re not helping anyone if they don’t have working plumbing.

The discussion veers into relatively standard discussions of AI doomsaying.

At 39:00,

“Gawdat: The only way we could reset is by resetting the entire Internet.
James: Now, is that something that could ever happen?
Gawdat: Never. I was sitting in silence the other day, and I wrote down three quadrants…”

JFC. This is definitely the wrong interlocutor for Gawdat. Somebody needs to call him on his sweeping, bullshit statements. “Reset the Internet” “1 Billion Times Smarter”. C’mon. This is kind of fun, but it’s not a serious discussion, because only Gawdat is contributing to this discussion. He’s now spending a ton of time explaining how people are selfish and incapable of working together above a clan level. Duh. Or that no-one can really say where the Internet actually is, or where it is. Interesting question, but he skips away quickly to talk about how awesome intelligence is.

He just can’t stop.

“Gawdat: I tend to believe that abundance of intelligence normally uh you know is correlated to abundance of ethics.
James: [nods vigorously]”

What? You’ve got to be kidding me. The relationship is nearly inversely proportional, with a few outliers.

If you thought he wasn’t going to double down on that statement, don’t worry.

“So, […] the dumbest of all of us would be destroying the planet […] and causing global climate change without even being aware of it you know. The less dumb would be destroying the planet despite being aware of it. Then, the slightly smarter will attempt to stop destroying the planet because they’re aware of it. The smarter still would attempt to fix the planet because they’re aware of the damage right, and you continue that trajectory. The smartest of all will always be pro-life. I always say that human arrogance makes us think that we are the smartest human—smartest being—on the planet. That’s not true at all. The smartest being on the planet is life itself.”

I mean, ok, I guess? This is just one of those statements where you can make out of it what you want, but the thrust is that intelligent people are the only ones possessed of sufficient ethics to save us. Plato lässt grüssen.

James just says “I love that” to everything, but Mo doesn’t even notice that he’s basically just talking to himself for 90 minutes. This didn’t need to be an interview-format video, with two people. It’s like 50% of the video screen is just a reaction video of James’s goofy head.

At 50:40, James tries to ask a question,

“James: What kind of control and ownership do we have as individuals, over the power of … Gawdat: That’s the most beautiful question of all.”

He didn’t even let him finish asking the question! He instead shoots right back into talking about a book he wrote (Scary Smart, as he’s done several times already).

At about 53:00 or so, he launches into a discussion of ethics, absolutely confusing social mores with ethics by giving an example of a Brazilian girl in a G-String versus a more conservative girl in a Muslim society. They are both respected for doing the right thing in their society, I guess? Those are just cultural habits.

I would have focused more on the underpinnings that led to those behaviors, like whether women have the same autonomy as men. But, yes, ethics is how societies resolve moral questions, like what is good, virtuous, evil, so I guess it fits. And he gets to say “G-String” and summon up the image of an ample, bronzed, Brazilian booty.

This whole section is about bias, but he thinks we can control “the ethical code of that machine.” Which, if he’s right, then it’s already too late, no? The machines have been built with the “wrong” ethics.

Then he hand-waves some stuff about how governments will have to build their own AIs to prevent AIs from being used for evil, then shoots right past that to give examples of how enough swipes on Instagram can help fix the ethics of an AI. Whooooooo. This guy doesn’t know many people. Has he heard of Internet trolls?

But then, but then, but then, he complains—for what feels like the fourth or fifth time—about people on his social-media accounts who are mean to him, when all he wants is to make billions of people happy. My cult-leader spidey-sense is going off to beat the band. And James is just nodding away like a dashboard bobblehead on a bumpy road, while the top comment on the video is “[h]e is down to earth.”

What is happening!?!

I think Gawdat could be so much of a better person if he didn’t spend so much time interacting with idiots online. Then, maybe, he wouldn’t have to make 40-day retreats to get right again. I see it in many other people I follow: otherwise intelligent people who end up making the broadest comparisons and most-shallow and incorrect arguments, just because that’s how they’ve been taught to think by the kindergarten schoolyard that is online discourse.

I was just listening to the Useful Idiots Podcast, with Aaron Maté and Katie Halper. I really like them. I think they’re intelligent, witty, and have their ethics in the right place. But they drew several conclusions that were absolutely the correct ones, but justified them with completely specious reasoning.

It’s the kind of thing that makes you so assailable. You don’t lock down your point because you made it in a way that someone who’s looking to disagree with you, no matter what, is going to be able to use to continue the discussion long after it should have been shut down. I think that’s my problem with Gawdat as well—his interactions have encouraged him to be lazy in his justifications for what I agree are the correct sentiments, which means I can’t really use anything he says as ammunition. It’s a pity.

At 01:05:00, he argues for the essential goodness of humanity,

“Are there more serial killers in the world or people who condemn killing?”

Sure, there are more pulses who are essentially good. Fine. Correct. But it’s the assholes who seem to have the overwhelming share of power and influence. The essentially good don’t have any influence. Jesus was wrong. The meek aren’t really lined up to inherit shit.

He touches on this as well, saying that the worst people are in politics, who get all the money, who are contributing the most information to the AIs. He says “the best of us” have “a duty” to take part. Sigh. Who’s the best of us? Which ethics? Implicit in his line of reasoning is that there is such a thing as “good ethics”, else with what would you align an AI? How would you select the “right” people for politics and training AIs? Plato’s philosopher kings all over again.

“You can’t succeed by being good. And it’s the most important time in human history to be good.”

He dances around the topic of how the system is utterly broken—perhaps because it’s how he even got to a position where he has more money than any human needs and everyone wants to know what he has to say.

When James asks him whether anyone can just ignore AI, Gawdat cuts him off again, saying “you will die in two or three years.” Wait, what? Then he clarifies,

“As a business. It’s as if you were trying to hang onto the fax machine in the age of the Internet.”

I’m sure everyone’s getting tired of me picking Mo’s nits, but he really, really elides so much in his analysis of “the world.” He uses “the world” as shorthand for “all of the 1%-ers he knows in Silicon Valley will have to adopt AI or their businesses will die.”

Most of the world doesn’t have use cases for AI, but he doesn’t think of them—or he’s deluded into thinking that they do have use cases somehow—or that they can be convinced to have them. He whipsaws back and forth between talking about his extraordinary empathy for his fellow man—and his utter inability to understand that the things that make humanity worth preserving have nothing to do with electronic mediation—or with the coming AI mediation of interaction.

He speaks very quickly, but I get the distinct feeling that he’s very wide, but not very deep. He is what passes for deep in his circles. But he doesn’t really know any hoi polloi. Nor does he see that as necessary. They’re the “dumb” ones “without ethics.” He’s slicker, but this is the basic line of reasoning of Hillary Clinton and her ilk.

He values intelligence above all else. Nothing else even comes close. That’s not how the world works. Everything is important. Intelligence can be leveraged. But intelligence doesn’t fix the indoor plumbing. He sounds kind of naive, but I think his spiel is also perfect for telling billionaires exactly what they want to hear.

Hell, they could be getting worse advice, don’t get me wrong, but his advice is so suffused with that hustler mentality—“whatever job you’re going to choose, choose the job where you’re going to be in the top two of people [who] can do that job”—all while he won’t shut up about silence and retreats and mediation and spiritualism. Really? The TOP TWO? Like, does that mean you shouldn’t work at McDonald’s? Who are you talking to, man? Like just your circle of self-selected …. philosopher kings. And every idiot in his cult will think “he’s talking right to me!”

Then he corrects himself to say “2 out of 10”. “Whatever you do, choose a job that you’re very good at.” James: “That’s powerful” Christ almighty, James, you’re terrible. Meanwhile, 90% of the world is just looking at Mo, going, “choose” a job? Luxury!

At 1:20:30, he says.

“Steve Jobs was successful because he had an empathy for the user’s needs, an appreciation of beauty, and enormous creativity—that actually are all feminine qualities.”

There he goes again, with his masculine and feminine qualities. Am I missing something? Is this not junk science? How does he get away with this kind of talk in his circles?

At 1:23:00, James says “I want to ask one last question.” Dude, did you even get in a first question? I’ve just been watching your nodding head in the left-hand-side panel like you’d been generated by NVidia’s AI.

In fairness, I liked part of Mo’s answer, describing what he thinks “purpose” is.

“I think the definition of purpose as per the Western society is very much commoditized—it’s almost like a target. It’s like, I set a target in the future. I spend the next eight years pursuing it, feeling frustrated and upset that I haven’t achieved it and then. when I achieve it, I have one of two choices.

“Either to set another target and feel upset for the next eight to nine years or to feel empty and feel that I’m purposeless. That’s a very misleading view of purpose honestly. It’s a very misleading view of the game of life in general.

“Because the only point in life that you have access to is right now. The Eastern philosophies will tell you: no, how can you set your life around the future, centric moment when life is here and now? How can you do that? The only way you can actually live life is to live here and now and so the definition of purpose becomes very different.”

Why would he think people would “hate him” for that? Ah, because he knows his audience is full of high-optimizing tech bros who are interested in appearing deep, but are really interested in money, and funding, and retiring.

“The purpose of life is to become the best you can be at something that you want to be and that makes life better for others.

“If you define life’s purpose this way, it becomes so easy. Because you know what
the one thing that a writer can do to achieve that purpose? It’s to write. Even if what you write is discarded, the purpose is not the book that I’m writing. The purpose is to write.

“That way of looking at life is very different than the Western way and I think that way of looking at life—’I want to become the best at whatever it is that I can do’—that is the right way to live with purpose.”

He keeps talking from the viewpoint of something who’s achieved a lot and who is very intelligent, constantly making the assumption that everyone else can achieve like him. Or, if not, not addressing the reality that most people who achieve the best that they can be at something are not going to be able to support themselves in the world we have.

The world we have doesn’t support this type of purpose for more than 5% of the people. We should have such a world, but we don’t. Yet. If I were in James’s place, I would have pumped him much more for ideas about how he thinks we can get there from here. How can we make the person who cleans toilets feel like they’re valued, like they’re living their best life? I’m not kidding. This is the problem you would need to solve.

It’s a shame that James just yes-manned his way through the interview because I feel that there’s much more there—or maybe we would find out that there isn’t. The other interview I saw with Mo Gawdat was very much in the same style.

At the end, Mo says “this was a wonderful conversation. At least for me, I felt it was really connected and deep.” AHAHAHA. He spoke for 99% of the time. He was talking to himself, pretty much. He’s not lying. He had a great time. James gave himself whiplash nodding for 90 minutes. 😉

A friend with whom I’ve discussed AI several times—among other topics—recommended the podcast So You Want to Be a Sorcerer in the Age of Mythic Powers… (The AI Episode) by Joshua Michael Schrei (The Emerald). I liked it very much. The entire episode is good; my notes and transcript start at just over an hour in.

The presenter’s voice is soothing, even if his cadence seems, at times, a bit forced. Overall, the effect is good. It was kind of ironic when he said that perhaps, in the future, people wouldn’t be able to tell whether a podcast was being read by an AI. He didn’t notice the irony, but I did.

At about 1:05:00, he sums up a longer section on the damages caused by modern humanity, failed experiments in social engineering that always seem to miss their espoused goals, while enriching an elite,

“Modernity is humanity seeing what it can get away with.”

Humanity is capable of much but, of late, it seems mostly interested in get-rich-quick, effortless scams.

At about 1:10:00, he expands on the topic of patience, on how it’s not a coincidence that the most revered figures in our most enduring stories and myths exhibit patience, are made to exhibit patience as part of their initiation to wisdom.

“In the stories, the young initiate who wants to access formidable powers, has to do what?

“Wait.

“You’ve seen the movies, you heard the stories, right? Of the master making the potential disciple wait outside the temple gate?

“You want access to the great powers? You’ve got to earn it.

“And the first way to earn it, before any physical trials, before any tests that take the would-be apprentice to the brink, the first way to earn it is—to wait.

“You’ve got to know how to wait.

“You know what the very first step of mystery-school initiation often is?

“Silence.

“The ability to sit with what is, without altering it, for a long period of time.”

This concept of patience—of earning—is, of course, wholly incompatible with our society, especially with the self-proclaimed elites who want to lead us off the precipice in their fervent hope that they will benefit in some short-term and frivolous way that is considered valuable by the short-term and frivolous society that somehow manages to buoy these selfsame elites on the backs of people so much more useful than they.

Patience is a virtue.

There’s a whole, incredibly soothing section where he convinces me that I’m a duck. Immagonna just leave it at that. I didn’t hate it.

At 01:32:00, he talks about the scene in the Matrix where Neo “learns” Kung Fu.

“It’s an awesome scene, right? And, of course, anyone who’s studied Kung-Fu—or any other somatic art—also knows that it’s a laughable scene because, simply, that’s not how bodies learn. Bodies learn through the time it takes to weave things into tissues. Bodies learn as patterns seep into the seven datus, the seven layers. Learning, knowledge, is an endeavor of bone marrow, and blood, and sweat, and breath, and proprioceptive weaving, over time.”

After doing some “like causes like” examples (e.g., if you want it to rain, than you ritually pour water, … um, … OK), at 01:39:00, he says,

“This daemonic power is not neutral. It is not a neutral intelligence that is being called up.

“By choosing which aspects of the living web of intelligence are the valuable intelligences and which are not, it is already value-laden.

“By centering rational empiricism, it is already value-laden.

“By removing intelligence from a body, it is already deeply value-laden.

“That is a value statement.

“By making it irreligious, aspiritual, it is already value-laden.

“AI is a biased God.

“Talking to ChatGPT, for example, is nothing like talking to an Aboriginal elder.

“It’s more like talking to a Stanford computer-science grad with an incredible analytic capability and very few real-life social skills. We are taking the narrow, world-naive, uninitiated, unembodied intelligence of the eager, neoliberal, Stanford grad and magnifying it on a global scale.

“Just what the world needs, right? All the biases inherent in the Western, scientific, analytic view of creation that has already taken us to the brink of eco-collapse—magnified 10,000 times.”

Goddamn, we need more philosophers, anthropologists, sociologists, historians, and linguists helping us run the world.

At 1:43:00, he wraps up with,

“[…] morality can’t be programmed in.

“Ethics can’t be programmed in.

“It can’t be programmed into machines or into human beings. [1]

“For all the current necessity that there is for ethical regulations, moratoriums, waiting periods, before the rush to market—these are still surface measures.

“When will we realize that trying to add ethics, […] to a system that is by nature hubristic, that is by nature at odds with the Gods, isn’t a viable long-term solution.

“Within the soulless fragmentation of late-stage capitalism, in which all things are pillaged and sold, and it’s everyone for themselves, all of the time.”

The post Now add a walrus: Prompt engineering in DALL-E 3 by Simon Willison is a story about someone gaslighting himself into believing that LLMs work better than they do.

Case study: pelicans and walruses

Willison prompts “A super posh pelican with a monocle watching the Monaco F1” and gets the following ideas.

So far, so good. It’s really wonderful that you can get something that’s not completely random garbage. However, the bird is only watching the race in the top-right picture. In the first and fourth, it’s definitely facing the fourth wall. It seems to be posh in all of the pictures, to one degree or another—indicated by a monocle or a bowtie or both. The first prompt asks for a “Photo”, but that doesn’t look like a photo. Still, cars, coastline, pelican. OK.

Then he says “More like the first one please”:

Look, I get what he’s done here. He’s trying to show how cool it is that you can make a ‘conversation” out of this by implicitly referencing an image that was in the response to a prior question. This didn’t used to work at all, less than a year ago, but not billions of dollars, thousands of developers, and millions of GPUs have made it possible. Kudos.

I guess the LLM interpreted that it should stick the monocle because the bowties are gone now. Willison is over the moon about how he thinks that it really got what he meant, but … the three new pictures look a lot more like the second picture than the first one (which features the whole pelican). It’s still doing reasonably well but, if a human had produced this, you’d be pretty annoyed that it’s wasting your time. It didn’t understand what you wanted and just made more pictures, but not “more pictures like the first one.”

Next up is “Add a walrus.”

In response, he writes that “[t]hat second one is amazing. [emphasis in original]” Does he mean the one where the walrus is photo-bombed into the foreground? That’s not really amazing, is it? The walrus isn’t watching, but neither is the pelican—but he didn’t ask it to make the walrus “watch”, just to “add” one, which is, I guess, exactly what the LLM gave him. The last one looks nice, but they’re not watching the race at all (just “attending”?), and the background contains speedboats instead of F1 cars. In the third one, the F1 car is in the water, but that’s OK, I guess?

Lowering your expectations

He continues playing with it, and being amazed at how it manages to kind of respond to his input, but shouldn’t we expect better? Maybe he’s amazed that it works at all, but we’ve got to get a bit more critical of this stuff—otherwise, it will continue to just generate medicocre images that only vaguely fulfill the requirements.

It’s the difference between asking a child, an apprentice, or a professional painter for a picture of a tree. You wouldn’t be at all satisfied with the output of a child from an apprentice, nor with that of an apprentice from a professional. I suppose my expectations are higher.

Missing fidelity

I completely agree that the LLM is able to respond to commands, but it’s not useful yet because it’s not able to make a finished product for you. You would have to tweak it to fix it.

And here’s the crucial difference between image-generation and text- or code-generation: it’s really, really hard to tweak the rendered image. Even if you knew how to use vector- or photo-manipulation tools, DALL-E is delivering a completed product, not the source that you would need in order to tweak it further. There are no layers in there. There are no masks. It’s just pixels. It’s a dead-end.

With text, on the other hand, we at least have the possibility of refining it in an editor. The finished product is itself editable at a fine-grained level. It’s entirely possible that you won’t be able to refine the product because you either don’t understand the language in which it’s written or perhaps because you couldn’t have done better yourself (which kind of amounts to the same thing).

I tend to think of code the same as I think of text: for a large number of languages, I can refine it better than the LLM could. If it’s a language or runtime library I’m not familiar with, or not well-versed in, then I may also not be able to “fix it up”, either.

This is the situation that most people find themselves in with code, and in which we all find ourselves with images. Even graphics artists can’t manipulate the output of an image generator, whereas text or code output could conceivably be improved by somebody, even if it’s not the person who prompted the LLM.

If someone claims to have seen a nude of you, but no-one can find it, does it exist? The article Teen boys use AI to make fake nudes of classmates, sparking police probe by Ashley Belanger (Ars Technica) should be addressing the question, but doesn’t. [1]

“According to an email that the WSJ reviewed from Westfield High School principal Mary Asfendis, the school “believed” that the images had been deleted and were no longer in circulation among students.”

But it also sounds like the school “believed” that the images even existed in the first place. Nobody reliable has ever claimed to have seen them—just teen boys, who are notoriously unreliable.

Back in the day…

Hell, when I was in high school, I might have claimed I’d made naked pictures of girls in school, just to fuck with everybody. I know my friends and I would almost certainly have thought it was pretty funny. we have even done it, just because people thought that we shouldn’t, but couldn’t stop us. We might have kidded ourselves into believing that we did it to remind people what free speech is.

I don’t think it’s that funny anymore, but not because I think people are justified in ruining their own lives worrying about it, but because I’m just generally unable to enjoy certain things I used to enjoy immensely, perhaps because of a reduction in ignorance, perhaps on account of a general reduction of being able to enjoy things at all. Overthinking has its price.

If someone claimed that they had a naked photo of me, it might make me uncomfortable—or it might’ve made me uncomfortable—but if the simultaneous claim was that it had been generated, then what the hell am I going to do about that? If I’d given out a photo of myself under certain conditions—perhaps implied conditions—then I have a right to be mad about those conditions being broken, but I certainly don’t think that there’s any legal recourse. If there weren’t even implied conditions—to say nothing of explicit ones—then what the hell am I going to do about it?

What I mean is: how could it be wrong to just say you’d seen something like that? It’s not even really conceivable that it’s illegal to have a naked picture that you made and then you say it’s a girl in school. What if you were really good with a pencil, and you drew one of them? Is that illegal?

Get a fucking grip, people. You can’t legislate this kind of thing and call yourself a free society.

The hot take

“It remains unclear how many students were harmed.”

No-one! No-one can even confirm that there are pictures, other than the say-so of a bunch of teenage boys. I’m not being a dick about this; read this summary,

“The school had not confirmed whether faculty had reviewed the images, seemingly only notifying the female students allegedly targeted when they were identified by boys claiming to have seen the images.”

Oh, man, am I glad that my anti-authoritarian self grew up in a world where you couldn’t get thrown out of school, to say nothing of being prosecuted, for saying that you’d seen salacious material about real-life people, just for fun. Talk about an entire society that can’t take a joke.

“Some of the girls targeted told the WSJ that they were not comfortable attending school with boys who created the images. They’re also afraid that the images may reappear at a future point and create more damage, either professionally, academically, or socially. Others have said the experience has changed how they think about posting online.”

My immediate reaction is: Oh, man, listen to that lovely language. “Not comfortable” … then throw them out of school! Might “create more damage” … how can fake pictures of you create more damage? We have to create a world where people dismiss this kind of shit—it’s not going to stop. Maybe we should make naked, porn-posed pictures of everyone. And then there’s the possibility that it’s “changed how they think about posting online” … Good! You should be thinking about what the hell you’re posting online, you goddamned narcissist.

Ok, let’s take a crack at a non-immediate reaction and see where we land.

Feeling unsafe

Instead of just being delighted that they’ve gotten to a place where they no longer have to worry about being actually being harmed, where it’s no longer a real concern, many elites and their children now commonly confuse “being made uncomfortable” with “being harmed” and decided to root that out, no matter what collateral damage is done to rights and free expression.

As detailed in the article, they’re trying to move us mentally to a place where we think that it’s OK for someone to get in trouble at school for saying that they’d seen a picture, that someone else had created, of a girl, depicting her in a way that makes her uncomfortable—or, in the vernacular of the day, makes her feel “unsafe”.

The story might be made up, the picture might not exist, the picture might not be of her, it might be a simulation of her, it might be based on a picture she’d uploaded herself. Now she feels unsafe. The world is a less-safe place for her because she now thinks that everyone is picturing her naked, or thinks she’s a slut, or whatever.

But this feeling of being unsafe could arise wholly without anyone actually doing anything at all. One of her friends could tell her that someone had told her that they’d heard that other people were trading AI-generated pictures of a naked woman with her face on it. She might think she heard one of her friends could tell her that someone had told her that they’d heard that other people were trading AI-generated pictures of a naked woman with her face on it.

It doesn’t matter how it happened—the feeling can be created in many ways, some real, and some fantastical. In all cases, that feeling is real, to her. In which of these cases is this feeling actionable, though? Should any of them be actionable? How realistic is the goal of preventing the world from ever making her feel unsafe?

It’s completely unrealistic. You can’t plug all of the holes [2] in someone’s fantasy. And what if the girl deliberately invents her feeling of being unsafe? What’s to prevent her from doing so? Hell, if she invents it well enough, it can even become real to her. Which, as we’ve discussed above, is the same as all of the other methods by which she could come by a feeling of being unsafe. It’s the feeling that society is trying to prevent, regardless of the path one took to get to that feeling.

This is how the mind works, especially for a teens, a group that is notoriously highly susceptible to peer pressure. If you’re the kind of person who feels mortified because you think people have seen you naked, you’re going to find some way of being mortified anyway, no matter how many fingers society shoves into the dike. [3] It’s just a matter of time.

We have to do something!

Do we, as a society, want to encourage this kind of mortification by helping these people punish other people for perhaps having inspired it? Are we so sure that we can tell the difference between all of the ways in which this feeling of insecurity/unsafeness could be engendered?

Of course not. We’ll draw an arbitrary line protecting the squeakiest wheels and damning less-squeaky or less-relevant people to punishment that they might not even deserve.

This is how we used to treat girls. Now it’s how we treat boys. The victims have changed, but the policy is the same: take the path of least resistance to protect your career and position in society, generally by shitting on the less-powerful people in any power dynamic.

Generally, you’re not going to get fired these days for punishing boys for looking at AI-generated naked pictures of their classmates. You might very well get fired for not doing anything about it, arguing, as I have above, that there is nothing you can sensibly or morally do about it, really.

In the old days, we’d have said “boys will be boys” and that was very wrong and stupid. We would say “boys will be boys” when those boys had actually harmed people. Now we do the same thing, but abet girls in meting out harm against boys, for a perceived harm that can’t be proven. Neither of these cases was or is OK.

I’d sent the post Somewhere in America there is an absolute legend who writes ‘SLUTS’ on box cars in various styles (Reddit) to a friend. He wrote back that they were “majestic sluts indeed”. I realized that I’d finally found a prompt to throw an LLM’s way. So I headed over to Stable Diffusion and prompted it with “Majestic sluts in the style of Boris Vallejo or Frank Frazetta” and chose a style of sai-fantasy art not because I knew what I was doing, but because I figured I’d give it the best shot I could. It responded with the following image.

Majestic Slut in the style of Boris Vallejo or Frank Frazetta (generated by Stable Diffusion)

Ok, so let’s analyze that.

• ✅ The LLM didn’t refuse to process my prompt because it had the word “slut” in it.
• ⚠️ The color palette is pretty close, but a bit too happy? Frazetta was darker.
• ⛔️ It assumed that a slut was female (most likely because of a ridiculous and discriminatory preponderance in the training data).
• ⛔️ The face is OK, but not really evocative of either of the artist’s styles.
• ⛔️ The pose is very generic and also not sufficiently contorted to evoke either of the two masters’ work.
• ⛔️ The breasts are porn-star breasts, not Vallejo breasts. Vallejo understood that breasts do not defy gravity
• ⛔️ Ditto for the fundament.
• ⛔️ The feet are Barbie-doll feet, posed for high heels, not for springing on a dragon’s back.
• ⛔️ That outfit looks more like lingerie than fantasy mail-armor. No dangly bits.
• ⛔️ There’s no sword, no tiara, no chain-mail bra, no dragon, nothing.
• ⛔️ There’s only one person in the image, when I very clearly wrote “sluts”

So, what’s the conclusion? Well, it’s in the ballpark, but I pretty much put it there by naming two of the artists from which it should draw inspiration. Also, I chose the sai-fantasy art style to seal the deal. From those parameters, even a web search would have found thousands of images from which to produce something.

To be honest, this image has probably been generated millions of times already by the long-suffering LLM at Stable Diffusion, which probably has to render “HAWT GRRLLL” for 99.9% of its prompts. Can you even imagine how many horny teens are trying to generate their fantasy girl instead of learning how to draw? [1]

I only threw one prompt the machine’s way. It was kind of close, but not good enough to use. According to Images that Bing Image Creator won’t create by Stewart Baker (Reason), this is a typical experience.

“As always, Bing’s first attempt was surprisingly good, but flawed, and getting a useable version required dozens of edits of the prompt. None of the images were quite right.”

That article is about the trust and safety limits that prevent certain content from being created in the first place.

“This is almost certainly the future of AI trust and safety limits. It will start with overbroad rules written to satisfy left-leaning critics of Silicon Valley. Then those overbroad rules will be further broadened by hidden code written to block many perfectly compliant prompts just to ensure that it blocks a handful of noncompliant prompts.”

That’s one concern, of course. Mine is more that we’re going to be satisfied with the absolute lowest-common-denominator of answers and recommendations and interaction.

It’s already been pretty bad, no? When you search for “horror movies” to find out what to watch, you just get a short list of the horror movies that everyone else is watching. You used to do it by selecting “movies in the last 365 days” and “horror genre” and “box office” and returning that. Great. Now, we’re doing the exact same thing, but with AI (LLMs).

Sure, it seems to understand a natural-language query; sure, it delivers a nicely formatted, natural-language result. But it’s the same list as before. Big deal. I saw some of Opera’s promotional materials for its in-browser Aria service. that featured a prompt that asked for three movies, but Aria delivered five. Even the people trying to sell this stuff don’t notice that it gets the easiest stuff wrong. 🤷‍♀️

LLMs are just a fancier way of getting you to consume mainstream, generic content. As usual. There will be no pleasant surprises, no growth.

The ~23-minute video below isn’t that long, but it packs a lot of information. The interviewer is insufferable, but Meredith Whittaker (president of Signal) is a force of nature, and Frances Haugen is very good, as well.

The Futurist Summit: Lessons of the Last Decade by Washington Post Live (YouTube)

At 08:00, Whittaker talks about the recent strikes in Hollywood,

“[…r]egulating AI, just non-traditionally. They did the classic move—withholding their labor—and they got terms that are actually staunching the bleeding of the use by the studios and big tech to place AI within their labor process that will degrade their labor, that will degrade artistic output, and will have a precedent-setting move of stopping the real harms, right now. I would look to the Writer’s Guild of America, I would look to SAG, I would look to your driver’s unions that are contesting the sort-of automated precarity of systems like Uber and Lyft, I would look to sort-of movements from below that are actually tackling the harms now, and not simply sitting around taking selfies with Elon Musk and calling it a regulatory agenda.”

Frances Haugen is also very, very good. At 09:50, she says,

“There is a skills escalator. You know, you come out of college, you come out of high school, and you have relatively low-complexity jobs. I had lunch with a friend a couple of days ago, and she’d been playing around with generative AI. And she’s like, ‘I’m never gonna hire a junior copywriter again! It’s like amazing!’ and I looked at her and I said ‘Amazing for you.’ Right? In a world where you’re a junior [list of jobs] … the jobs that allow you to become a more sophisticated contributor—they’re about to disappear.”

The dipshit interviewer responds with “clearly, yes, there is going to be huge impact on labor.”

No. Jesus, lady. Could you be any more indoctrinated? Can’t you hear what Haugen is saying? Even if she were wrong, you should still, as the interviewer, engage her argument, rather than blowing right through to your predefined agenda. No wonder Whittaker keeps rolling her eyes. Do your job. Actually, you know what? At least she shut up and let Haugen speak her piece.

What Haugen is pointing out is that the already pitiful “training program” that the U.S. has is going to become even worse—it will be utterly broken. Businesses only ever put up with having less-skilled employees around because they were investing in them to become more-skilled employees. If AI replaces less-skilled employees, there will no longer be more-skilled employees either—because where will they come from?

The U.S. already lacks a training programs for so-called blue-collar jobs. Now it’s going to wipe out its ad-hoc training programs for white-collar jobs. At least places like Switzerland still have apprenticeship programs. We’ll see how long that lasts, though, as every so-called advanced country chases the U.S. down the drain.

Whittaker is devastatingly insightful. She draws the distinction between an actually useful technology and the “bombast” surrounding it, delineating that the problem is with the hyper-capitalist companies that own and drive the technology—“it’s the definition of metastatis”—rather than with the technology itself.

At 22:40. she says,

“Just to clarify: ‘hype’ doesn’t mean it doesn’t do some things. Hype means that an entire ecology of narrative bombast has been predicated on … yeah, it can help you write an e-mail. If that’s a problem you want to solve with 20 billion GPUs, you can do it. But is that a world-changing problem to solve? And what is the actual material basis for what I would call these bombastic claims? […] Let’s get back down to reality and the actual the thing it [GPT] does before we make all of these predications based on that.”

The point of the bombast is to increase stock price in the short-term. There is literally no other goal anymore. Maybe there never was.

AI is going to ruin everything. No. Capitalism ruins everything.

The tools are useful, but the companies that own them are willing to lie about them in order to make them seem more useful to everyone. They sell an Eierlegende Wollmilchsau and the same fools lap it all up, just like they do every time.

It’s the opposite with vaccines. We’ve not had a single technology that has helped save more lives in the history of mankind. And yet, vaccines have never had a worse reputation than they do now. People don’t trust them. They don’t think they work. It’s a clusterfuck. And that has a lot to do with the way the hyper-capitalist system has benefitted from vaccines.

Instead of imagining that we could get inexpensive, reliable vaccines for everyone, we accept that they will always become more expensive as the companies that control them tighten the noose. We accept that we never will wrest control of vaccines from these companies, so we write them off instead! The most effective medicine ever—and we choose to ignore them rather than to imagine controlling them ourselves.

It really is true that it is easier to imagine the end of the world than the end of capitalism.

The discussion on Yann LeCun: AI one-percenters seizing power forever is real doomsday scenario (Hacker News) also has several good comments about AI and capitalism.

A co-worker of mine sent me the following video with a strong recommendation. There are parts I liked, and parts I did not. It was a long video. The following are my notes on it. My attitude starts off pretty bad and gradually improves, then goes a bit downhill again.

YouTube

These people are all fools or shysters. The young guy (Stephen Bartlett?) interviewing offers as proof that AI is amazing is that his miniscule mind is already satisfied with it. *applause*

The older guy seems like the kind of guy who’s been smart his whole life and has developed an incredible inability to conceive of a world in which he could ever be wrong. He flatters the host by calling him one of the most intelligent people he’s ever met. What in God’s name is happening?

They seem to be on track to trying to convince the world that two geniuses agree that ChatGPT is the way to go.

Gawdat says at 33:15 that he could have ChatGPT write a book for him.

“The only reason why I might not want to follow that path is because, you know what? I’m not interested. I’m not interested to continue to compete in this capitalist world. As a human, I’ve made up my mind a long time ago that I will want less and less and less in my life.”

It’s a nice sentiment, but it’s also spoken by someone who’s rich beyond all of his desires. He doesn’t need to compete anymore because he’s already won.

This video is two multimillionaires having a two-hour conversation, massaging each other’s egos and not really saying anything new or interesting.

If AI can ruin our culture and society, it just means that we built a dumpster fire in the first place. It means that we have a system that values people and humans so little that it would prefer to deal with whatever happens to be the first feasible simulacrum of a human. It will be like letting the prokaryotes take back over.

Gawdat at 41:00, expressing his anger.

“We fucked up. We always said ‘don’t put them on the open Internet. Don’t teach them to code. And don’t have agents working with them. Until we know what we’re putting out in the world. Until we find a way to make sure that they have our best interests in mind. Humanity’s stupidity is affecting people who’ve done nothing wrong. Our greed is affecting the innocent ones. The reality of the matter, Stephen, is that this is an arms race. It has no interest in what the average human gets out of it. Every line of code being written in AI today is to beat the other guy. It’s not to improve the life of the third party.”

Not “Humanity”, but the “self-selected elites”. Once again, capitalism ruins everything.

And he would go on to basically say that the problem is not AI or LLMs or whatever: it’s the system of capitalism we have, the system of society that we have, that is so zero-sum that we can’t think in any terms other than to “win”.

Win what? No-one can really say. People just want to be feel secure, to see how they will not become insecure unfairly, that they are appreciated and rewarded for participating usefully, that they are given a chance to be useful, that they are entertained, that they can interact socially. That’s it.

There is nothing in there that says that everything must be “bigger, better, faster, more” All. The. Damned. Time.

In fact, the faster things get, the less likely it is that most people will be fulfilled. People’s fulfillment is almost completely out of their hands right now. They don’t know what they want anymore.

They have been convinced to want things that require a tremendous machine to produce, a machine that, coincidentally, also transfers most of the world’s wealth to a paltry few hands while convincing the rest of the world not to revolt by producing a few shiny baubles and trinkets.

At 41:45, Gawdat again:

“And people will tell you that this is all for you. And look at the reactions of humans to AI. We’re either ignorant: people who will tell you, oh no no, this is not happening. AI will never be creative, it will never compose music—where are you living? You have the “kids” (I call them): you have them all over the Internet, they say ‘oh my God, it squeaks, look at it. It’s orange in color! Amazing! I can’t believe that AI can do this!’ We have snake-oil salesman, who are simply saying, ‘copy this. Put it in ChatGPT, then go to YouTube, knick that thingie, don’t respect copyright or intellectual property of anyone, place it in a video, and now you’re going to make $100 a day. Plus, we have these token evangelists: basically, people who say, ‘this is it; the world is going to end’. I don’t think that is going to happen. You have your token evangelists, who are saying, ‘oh we’re going to do this, we going to cure cancer.‘ Again, not a reality. And you have a very few people who are saying, ‘what are we going to do about it?’”

In fairness, it is composing and painting and producing text, but the bar is so low that it’s not really competing with human endeavors. What it is, though, is filling a massive gap that had traditionally been filled with mediocre human endeavor. That will be gone.

In that sense—even though it is still not conscious and not intelligent—our shitty system will imbue it with enough importance that it will allow most of what is good about society to be eroded away over night before we can even think of stopping it.

Our structures for living good lives will be gone. The only difference with this AI “revolution” is that it’s now affecting the self-important elites. The other 90% of the world has already had this happen to them during the first 45 years of neoliberalism.

Gawdat again:

“What went wrong in the 20th century? Interestingly, we have given too much power to people who didn’t assume the responsibility. […] We have disconnected power and responsibility.”

“I feel compassion for the rest of the world. I feel that this is wrong. I feel that for someone’s life to be affected by the actions of others, without have a say in how those actions should be, is the ultimate, is the top-level of stupidity from humans.”

He’s really just describing how the world works for 95% of the population, though. This isn’t to say what he’s saying is wrong, but that he’s saying it now because there is finally a real danger that the elites will be swept up in the madness that they sow every day. With this crop of LLMs, there is a real danger that money cannot protect you. That is frightening to the powers-that-be.

I think the more interesting things he has to say are about our underlying system, which makes the prospect of introducing something like even a half-functioning AI so much more … difficulty to handle with grace.

At 1:00:00,

“It is here. This is what drives me mad. It’s already here. It’s happening. We are all idiots, slaves to the Instagram recommendation engine.”

HAHAHAHAHA. Not all of us. Not even most of us. There are way too many people on this planet who are not dealing with this horseshit.

Just as an aside, though, he says that “70 years later, we are still struggling with the possibility of a nuclear war, because of the Russian threat of saying, ‘if you mess with me, I’m going to go nuclear.‘” This just goes to show how woefully brainwashed even intelligent people are about the real world, the stuff that really matters. He is an Egyptian. His first example of nuclear brinkmanship is Russia, not the U.S. It’s incredible. As he’s discussing how we’re all slaves to an algorithm, he shows how even his big brain has been enslaved by America propaganda.

The greatest trick the Devil ever pulled is convincing the world he didn’t exist.

A little later, Mo and Stephen make a few jokes about the evil Chinese and the evil North Koreans and how there would be no possibility for cooperation because of how evil those countries are. I’m shaking my head. These two are so in-the-tank ignorant about global politics and they think they can solve our problems for us? I shudder.

At 01:04:00, “They’re 1B times smarter than you.”

Um, Ok. Sure.

At 01:26:00, they discuss how to address this coming problem: their only solution is to work with the extremely restrictive incentives offered by the current system. I.e., what makes more money?

In fairness, this is most likely the correct way to approach the problem; we don’t have time to fix the system before we tackle the AIpocalypse, but, with the show clocking in at almost 2 hours, it would have been nice to acknowledge that the only reason their ensuing discussion is going to sound like a WSJ/conservative-think-tank/Silicon Valley startup round table is because we have to go to war with the army we have.

At 01:28:30, they talk about how international competition will always lead to other countries “letting it rip” with AI research/development, even if a country were to tax AI research/revenues in order to deal with the damage it causes. It’s the same as climate change.

Stephen says,

“It’s kind of like technology broadly; it’s kind of like what’s happened in Silicon Valley. There’ll be these senators who think that tax-efficient founders get good capital gains […] Portugal have said that there’s no tax on crypt … loads of my friends have got on a plane. And they’re building their crypto companies where there’s no tax.”

Hahahahaha. You should get better friends. Honestly.

He then bitches about GDPR as a failure because it’s “annoying”. Yeah, sure, if you just click away all of your data on every web site. The current implementation is a bit annoying, of course. But I’d rather have that than the alternative, which is that I don’t get any control over my data. The next step is to have the browser fill in GDPR automatically with your preferences: just as restrictive as possible, every time. Problem solved. Again, the problem here is parasites making money off of the CO2 that you produce.

At 01:43:00, Gawdat says,

“I don’t think we’ll be hiding from the machines; I think we’ll be hiding from what humans are doing with the machines. […] In the long term, when humans stop hurting humans because the machines are in charge, we’re all going to be fine.”

Sure, sure, OK. A bit of fully-automated luxury-Communist fantasizing. I’ll take it.

I'm with stupid ⬅️The article Does AI Just Suck? by Freddie deBoer (SubStack) writes, after providing two examples of a heavily feted AI utterly failing to create images of John Candy and Goldie Hawn, defaulting to middle-of-the-road “fat man” and “blonde woman” representations that leaves the viewer to fill in all of the gaps left by the mediocre effort.

From the essay,

“[…] you’d think that, among the various tasks you might charge an AI image generator with, recreating faces that have been photographed many thousands of times would be one of the easiest. What just drives me mental about this stuff is that tons of people insist on pretending that these technologies work as intended! In the thread where these images appear, there’s plenty of people who point out that they look nothing like their human counterparts, but also people going “Wow! Amazing!” That’s true of so much of AI-generated art; it feels like people have been told so relentlessly by the media that what we are choosing to call artificial intelligence is currently, right now, already amazing that they feel compelled to go along with it. But this isn’t amazing. It’s a demonstration of the profound limitations of these systems that people are choosing to see as a representation of their strengths.”

I agree with this impression. There are some things that look pretty wonderful, but it’s also hard to escape the conclusion that these LLM-based image-generators are good at creating generic artwork like the kind of stuff you’d have found on posters in a Spencer’s Gifts in the late 80s/early 90s. The essays feel like the output of a middle-schooler or shitty undergraduate who’s just trying to fill a page with text that feels vaguely relevant.

There is no spark of innovation—just a frisson that it didn’t completely fail, that the LLM got pretty close. We’re already amazed that it produced a blond woman or a fat man, even if it doesn’t come close to the representation that a reasonable artist could sketch in a few strokes from a handful of pictures—or even just one, as any halfway-competent caricaturist could do. Sure, it’s not just random pixels, but it’s also not really useful.

But what can it do for us?

As I’ve noted before, I also don’t see how we get to useful from here—precisely because we don’t know how it’s even getting a generic blonde woman from the prompt “Goldie Hawn”. I think it’s reasonable to ask whether these LLMs are the thing we should be prioritizing right now. For me, the answer is, clearly, no. But the Lords of capitalism have determined that they can mine some short-term value from it, so we’re stuck hearing about it until it suddenly implodes, washing away all value except that which has accreted to a handful of the richest people on the planet.

“As I will go on saying, all of this would be much lower stakes and less aggravating if people had the slightest impulse toward restraint and perspective. But our media continues its white-knuckled dedication to speaking about AI in only the most absurdly effusive terms, terms that threaten to exceed the power of language.”

This is the role of the corporate-owned media. They are an advertising arm masquerading behind a system that has the vestiges of gravitas left from a bygone age. Their role is not to inform; their role is to ensure short-term gain of value for their lords and masters. “[R]estraint and perspective” don’t enter into it, unless it would serve that goal, which it rarely does.

The fallacy of the online/offline knowledge base

“I’ve been telling people for a couple decades that the attitude of “kids these days don’t need to learn facts because they have Google” is fundamentally flawed, as learning facts is an indispensable part of creating the mental connections in your brain that drive reasoning.”

This is obvious to anyone who isn’t relieved to be able to offload all of their thinking to the online advertising and propaganda machine. You can’t draw conclusions if all of your knowledge is online. It’s not your knowledge—you can search for anything, but you have no idea what to search for. You are not an interesting conversation partner, you have no original ideas, you can’t innovate—because you don’t have any online knowledge.

Your processor might be powerful, but your memory banks are empty. Relying on LLMs for even more than we already rely on search engine for will only exacerbate this problem, will only lead to a world even more full of people who can’t reason their way out of being bamboozled by state propaganda. This is not a coincidence.

Building things well is difficult

“[…] what if this software just sucks? What if we’re all so desperate to move to the next era of human history that we talked ourselves into the idea that not-very-impressive predictive text and image compilers are The Future?”

That is entirely likely. Most software sucks. I find it hard to believe that software that has just appeared—has been grown, if you will—will be somehow better than software that actual developers have tried to design.

People somehow think that it’s better just because no-one understands how it does what it does. If you’re the kind of person who doesn’t understand how anything works, then you’ll like the mystery of it because literally everything else in your world moves in mysterious ways.

These kind of people don’t understand even 1% of how their world works. They don’t know where resources come from, where trash goes, how food can exist, how plumbing works, how any technology works—or why it doesn’t work or can’t work or why it might stop working—they don’t understand biological limitations, or how chemicals and pharmaceuticals are researched and developed, they don’t understand economics or politics or even basic social interations.

They find it reassuring that, with so-called AIs, no-one understands them. They can vaguely grasp that this means that, for once, they aren’t relatively stupid about a topic, as they are with everything else.

In the other cases named above, they have to assume that there are smarter people out there who do understand how things work—and that those people are better than they themselves are, that those people are more useful.

Those kinds of people are not reassured that we don’t understand how these LLMs do what they do—because they understand the scientific process, they understand engineering, whereby one has to understand what is going on, in order to improve it.

When you’re a blithering dolt who’s ignorant about everything, your approach to life is to just do stuff and hope for the best. There is no process. These LLMs are perfect for people like this. They already think these LLMs are amazing, mostly because of their ineffability, because it matches their own inability to grasp how anything works.

They’ll never notice that there is no predictable path forward for improvement in something that we don’t understand. But, in a country—heck, a world—addicted to gambling and ignorance, this fact won’t bother anyone. They’ll think that we can just blunder our way toward improvement, calling each change progress regardless of viability or usefulness to anyone.

The real reason is always the same

The only benchmark will be, as always, are the richest people getting richer because of it? If yes, then carry on; if no, then change course, regardless of utility to anyone else. Are resources wasted? Is energy wasted? Is effort wasted? Could the energy, effort, and resources have been invested more effectively elsewhere? None of that matters.

The only thing that matters is whether the handful of already-wealthy people and entities consolidate even more of generated human value unto themselves. Any other benefit is a side-effect. If that side-effect threatens the continued accumulation of capital? It will be reverted and avoided in the future. It’s why we can’t have nice things.

Hell, you can tell people that things are getting better and they will believe you—especially if you tell them often enough.

So, we’ll probably just stay with the current, shitty crop of LLMs that our lords and masters have dubbed “AIs”—and watch them get richer, while our lives approach the minimum quality that continues to deliver value upward while avoiding revolution.

As people get dumber and shittier and more egocentric as a consequence, the LLMs will actually start to seem more lifelike! So, we have that to look forward to.

The joke does not continue; my apologies. Unless the joke is that we will soon be even less able to comprehend, make sense of, or otherwise act on hypotheses about the world because we are accelerating our already advanced pollution of our information environment. What does that mean?

Actually salient information drowns in a sea of utterly meaningless noise. It’s been this way for a while, if you’ve been paying attention. Social media was the first booster rocket taking us further away from being able to influence our societies in any way that does anything to even think of negatively affecting the profits of our elites or the stranglehold they have over any and all levers of power.

You can see it in the shocking lack of information many people have about how the world works, or about any current events. But let’s go back to something with a bit more levity instead of focusing on the doom and gloom.

A friend sent me the site The Infinite Conversation, which is an AI-produced,

“[…] never-ending conversation between Bavarian director Werner Herzog and Slovenian philosopher Slavoj Žižek. When you open this website, you are taken to a random point in the dialogue. Every day a new segment of the conversation is added. New segments can be generated at a faster speed than what it takes to listen to them. In theory, this conversation could continue until the end of time.”

Part of the joke is that this AI product is so niche that it is utterly harmless. More people will recognize Herzog, but very few will recognize Žižek. I’m a fan of both [1].

It’s kind of sobering how realistic it sounds at first. You have to really follow along to tell that it doesn’t make much sense (i.e., mine had Herzog saying that his dream was to make films in the jungle; he’s already made several). And Žižek’s sentence wasn’t even grammatically correct – but that doesn’t mean it wasn’t Žižek. 🙃

That it makes no sense to the trained or expert ear is actually quite normal for AI-produced output. What AIs produce is usually kind of generic—like an undergraduate essay—or just outright incorrect. You kind of already have to know the answer if you want to be able to use it. People generally ask these things the kind of questions that it can’t hope to answer, but which they fervently hope will provide them with some sort of insight into how to proceed in their own lives. I see these AIs more as oracles or tea-leaf readers. You will get out of it what you interpret from the vague answers that they deliver.

We’ve always dealt with the possibility (and absolute reality) of fakeness in our information environment. It just used to be more difficult to produce it en masse. Now, we can spew out a literal TON of noise to signal, drowning out any hope of understanding our underlying physical reality even further. Now, when we would need most to understand what is happening, and now, when we are responsible for making decisions that will impact generations—if not the species—we are more befuddled than ever…and couldn’t really care less.

The largest misinformation campaigns go largely unignored, because they are official ones. There is the cult of Russiagate, which has poisoned nearly all thought not only in its country of origin, but it has also severely infected the mental hygiene of otherwise rational people in allied countries. That complete fabrications laid the groundwork for a renewed hot war with Russia is much, much more dangerous than these AI infractions, which are tiny in comparison (so far).

Eminence grise Donald Knuth documented his (indirect) interaction with ChatGPT (3.5, I believe) in a text file. His conclusion?

“I myself shall certainly continue to leave such research to others, and to devote my time to developing concepts that are authentic and trustworthy.”

Sounds good.

Drag-Drop Image Conversion by Simon Willison is a gist that contains the conversation that Simon Willison had with Ghat-GPT to build a drag-&-drop image converter.

First of all, he started working on it on April 1st, but it’s hard to believe that he’s pranking—he doesn’t seem the type—so I’ll give him the benefit of the doubt. Assuming that this is real, it’s impressive that it can turn those prompts into a working application.

Although … did it? If you copy/paste any of its examples into an HTML page, none of them does what it says they do.

The drag-&-drop doesn’t work. The output is just most of the code repeated in the text box.

It’s just impressive-looking and much closer than just random code, but they don’t work. If you don’t know how to program, can you fix it? Of course not.

The fixes Willison ended up making were very non-cosmetic.

Diff showing non-cosmetic changes

Still, let’s pretend that they had worked or gotten much closer to working. What if everyone could build code like this?

Hey, maybe it brings back the world of experimentation we had back in the early days of the web, when everyone was writing HTML directly. Maybe these simple things will start showing up more again, especially when a machine writes them.

I wonder, though, where it got this code from. Has it really built it mostly on its own? Or would we find a remarkably similar version somewhere in the vast input data that was its corpus?

Another question is: what are the implications if we would start building software this way? Do we just dump our notions of architecture and a common coding style? There are a lot of considerations about maintainability and consistency and onboarding and so on that go into our software today. Do we just throw all of that overboard and move to a patchwork of one-off components?

If we’re being honest, isn’t this how a lot of programmers are already building code for their employers? Just copy/pasting stuff together and crossing their fingers?

How to use AI to do practical stuff: A new guide by Ethan Mollick (One Useful Thing)

“You often need to have a lot of ideas to have good ideas. Not everyone is good at generating lots of ideas, but AI is very good at volume. Will all these ideas be good or even sane? Of course not. But they can spark further thinking on your part.”

I suppose this beats having friends or coworkers. Apparently the film “Her” was utopic, not dystopic.

“Summarize texts. I have pasted in numerous complex academic articles and asked it to summarize the results, and it does a good job!”

How the hell are you in a position to judge? You said before that it lies all the time, that it has no mechanism for admitting defeat because that doesn’t exist. It’s building text. It’s always successful. There’s no meaning to get wrong. It’s like reading tea leaves. The cup doesn’t know how to set up the leaves. The meaning is inferred solely by the reader.

If you don’t know what the paper is about, and you know the reputation of your tool to just make shit up, how can you possibly even think you can judge whether the summary it produced is reprentative?

“If you don’t check for hallucinations, it is possible that you could be taught something inaccurate. Use the AI as a jumping-off point for your own research, not as the final authority on anything. Also, if it isn’t connected to the internet, it will make stuff up.”

Hahahahaha sure. That’s exactly how a lazy, conspiracy-obsessed society treats technology and information. This guide actually applies to using the Internet in general, but almost nobody’s ever followed it. People just inhale information, with the only vetting process being “am I being entertained?”

Also, this is exactly the lesson he ignored above when he claimed that the AI did a good job of summarizing complex academic papers.

Schillace Laws of Semantic AI (Microsoft Learn)

“Don’t write code if the model can do it; the model will get better, but the code won’t.”

So treat the prompt like a high level language that targets a compiler that fabricates and whose workings we don’t understand. Interesting, so maybe just feed your requirements directly into the machine and hope for the best? At some point, it will come up with something that actually functions?

The code won’t get better on its own, but neither will it get worse. It will continue to do what it says on the tin. We may discover more negative ramifications, but what the code does will not change. The quality of the code produced by a prompt—or series of prompts—will change, but not necessarily only for the better, which is being strongly implied by this rule.

“Uncertainty is an exception throw. Because we are trading precision for leverage, we need to lean on interaction with the user when the model is uncertain about intent. Thus, when we have a nested set of prompts in a program, and one of them is uncertain in its result (“One possible way…”) the correct thing to do is the equivalent of an “exception throw” − propagate that uncertainty up the stack until a level that can either clarify or interact with the user.”

Understandable, but it sounds tedious and fraught. It’s getting farther from treating coding as an engineering discipline. Maybe something comes out of it—maybe it’s how everyone will be coding in ten years!—but it feels very wooey and very hypey right now. I can’t tell the difference between this technology and an actual scam, except that this technology kind of looks like it does something useful. It reminds me of a scam in some cities: you have people who pose as public-transportation workers who will sell you tickets. The tickets actually work. But they’re not valid for more than just the smallest zone. You’ll pay for five or six zones, but you can’t actually travel there. AI reminds me of that, so far.
 

Why Are Lithium Prices Collapsing? (Hacker News)

The comments are full of people heralding the growth of Lithium mining, as if there being more of it available has come at no cost to anyone. Of course, they don’t think about the destroyed environment or the destroyed communities—they think only of their privileged, 1% future because they know they only ever benefit from increased extraction—in the form of increased availability or lower prices or both—and they never suffer any of the ill effects. Then they flood the zone with comments about how the doom-saying Cassandras were wrong about limited resources because lookit how much there is naow.

The so-called Cassandras were never saying that there wasn’t more lithium. They were saying that the price for increasing production was high. That it would be wrong to extract more of these resources, if it comes at the cost of the environment or increased CO2-production. They have not been wrong.

Why are lithium prices collapsing? by Nichloas Larsen (International Banker)

I don’t really have to read the article to know the answer. Prices are collapsing because the most powerful elites in the world see the possibility of increased wealth and profits in selling every first-world idiot an electric car or three. Prices are collapsing because they will it—and because neither they nor anyone they know, or are aware of, will suffer any of the deleterious side-effects associated with increased extraction.

After spoon-feeding its audience a bunch of horseshit, it ends with a complete repudiation of the article’s title.

“And while substantial new sources of supply may come online eventually, some analysts do not see this having a sustained impact on prices in 2023. Scotiabank (Bank of Nova Scotia), for instance, believes the recent selloff in lithium equities has been largely unjustified for this reason. “While the year ahead has a slight chance to see temporary softness in lithium spot prices, beyond 2024, we are stumped as to where supply will come from to satisfy demand,” it recently noted, as quoted by the Financial Times. And Trafigura recently echoed this sentiment. “I really don’t think there’s any reason to believe that so many tons can magically appear this year to return the market to balance,” Claire Blanchelande, a lithium trader at the commodities trading firm, told Bloomberg in January. “The pain is not over yet.””

The post Please explain++ (Fuck Nestle) by PetronilaConaway (Reddit) is of the following image:

Winona LaDuke is 100% right

It reads,

“Someone needs to explain to me why wanting clean drinking water makes you an activist, and why proposing to destroy water with chemical warfare doesn’t make a corporation a terrorist.”

The top comment by Horse_Renoir reads,

“That’s easy!

“Capitalism expects infinite growth from our finite world so it is inevitable that capitalists will come for our every last natural resource.

“The entire system is setup to support capitalism. So if you get in the way of capitalists exploiting the things we need to survive in their quest to make a profit you’re a terrorist. 😡

“If you’re a capitalist exploiting the things peons need to survive you’re just doing as the system intended and are in fact a very good business person. 🤗”

I don’t think I’m an especially fussy user of software. I just can’t help noticing when it keeps doing stuff that it wants to do rather than what I want it to do. I also can’t help noticing how so much software manages to utterly fail to adequately do even the simplest tasks that are directly related to the thing they were built for doing.

🤦‍♂️ Apple Maps 🤦‍♂️

Today, I had Apple Maps open in Schaffhausen. I searched for a route from Winterthur Bahnhof to a restaurant. I left open for later. When I got to Winterthur, I opened Apple Maps again. I could see the route I’d planned, with a Go button. Before I could click the button, though, the app reloaded itself and asked me where I’d like to go. You know where the f*%k I want to go. I just had it on the screen before you needlessly refreshed. Why did it refresh? I’m sure that there’s a logic in there that says: if enough time has passed since the query or perhaps your location has changed enough, then just refresh the page. Why? Because there’s no refresh button, perhaps. But, it’s not smart enough to notice that I was literally at the starting point of the route that it so gracelessly erased.

🤦‍♂️ Mobile Browsers 🤦‍♂️

This is not uncommon on phones. Browsers generally like to reload themselves all the time, even when there’s absolutely no need to. This not only wastes data, but is often frustrating when you have no connection or only a flaky connection. In this case, you literally had the information you wanted on-screen, but then your gloriously intelligent supercomputer of a phone decided to throw this all away and try loading it again, in which case it just tells you, “whoops, sorry, that information’s not available.” Well, it was available, you exquisite dumbass, but you erased it needlessly. No-one asked you to erase it, and I could have told you that you wouldn’t be able to reload it because I can see that you have no data connection.

🤦‍♂️ The SBB App 🤦‍♂️

The SBB app is quite famous for doing this. Instead of simply showing the route I’d planned, it’s refreshing all of the time, even when it could very easily check that it has no data connection. Just. Stop. Doing. Stuff. Just show me the data that was on the screen. I will tell you when I want you to show something else.

The SBB app is also quite bad at linking its information. Today, I searched a route and had bought a ticket for it. I didn’t know where the route went after that, though. It shows the purchased ticket—or, rather, the QR-code for it—but then the itinerary that you’d purchased it gone. You can click on what looks like the itinerary, but it’s just the day-ticket that shows you the zones you’d purchased. Where’s the route? It’s in the “journeys” tab, which is ordinarily empty, or is filled with an advertisement for the “get-on-get-off” feature that they’re pushing so hard. I only checked there on a last-ditch hunch. It’s exactly what I was looking for. So, what’s the problem? Well, why wasn’t there a link from the ticket I’d purchased to the journey for which I’d purchased it? This is not rocket-science.

🎖️ Komoot 🎖️

One app that does a relatively good job with this is Komoot. That app kept showing me various levels of detail with its offline-map feature no matter how little data I had. It even worked fine before I’d purchased the offline-maps feature, in which case it was able to show me any of the zoom levels I’d loaded before I went off the grid. Very nice, and exactly what I expected.

🎖️ Garmin 🎖️

While on the subject of sports apps, I am shocked to be able to say that I was pleasantly surprised to see that Garmin’s algorithm handled by extra kilometers generously today. What does that mean? Well, I’d signed up for an expedition a while ago, and had only 1km of 162 left to go on it. I walked about 17.5km today, all together. Garmin awarded me the badge for the expedition, then transferred the remaining 16.5km to the next expedition that I signed up for! Nice! That was a pretty pleasant surprise.

🤦‍♂️ Apple Reminders & Calendar 🤦‍♂️

I’m looking at Apple’s reminders right now. If there’s a single one, you hover it and it’s supposed to show a little button that you can use to select “Complete” or “Snooze”. Sometimes it shows up; sometimes it doesn’t. If you click anywhere else in the notification, it opens the calendar or reminders app and you’ve missed your opportunity to complete or snooze it. Too bad for you. If there are multiple messages, then you can’t click the button. It doesn’t show up on hover. Then, you have to click somewhere in the notification, in which case it doesn’t open the calendar or reminder app, but instead unfolds the notifications to show them all individually. What are you training us for here, Apple? Do you even have a UX engineer working there?

🤦‍♂️ UPC Media Box 🤦‍♂️

I just turned on my UPC television. I was listening to a radio station on it last night. Did it turn back on to the radio station? Of course not. It defaulted to a arbitrary TV channel. It wasn’t even the TV channel I’d been watching before I’d switched to the radio. This is categorically and unequivocally awful. Now, I have to switch back to the radio, which is about seven clicks away. And why does UPC not remember which user was last selected? Every time I turn on the television, I have to select my user again so that the TV guide is a manageable size (favorites are associated with a user).

And the movie I was watching in French the other night? The one I have in my list to continue watching later? When I start it back up, the language has been reset to English—because, well, why not? Why would a computer in a box be able to remember a handful of settings along with the movie. At least it managed to remember where it was in the film. Thank God for small favors.

In all fairness, the show that the TV is forcing me to watch is Selby vs. Allen in the Snooker World Championships, so I’m not altogether angry about it.

🎖️ Apple TV 🎖️

In that regard, Apple TV is extremely good at picking up where it left off. If I turn it on in the morning, it picks up exactly where it was in the song that was playing the night before, in seconds. Good boy.

🤦‍♂️ Apple Mail 🤦‍♂️

On the other hand, my Apple Mail on my laptop shows 3 unread mails in a mailbox that has, very obviously, no unread mails. The badge on the app icon has shown first one, then two, now three, even when there are no unread mails. I see the flagged folder says “4”, but, when clicked, it shows only two mails. I’ve tried a few online guides to remedy this, but nothing seems to help.

🤦‍♂️ Apple Photos 🤦‍♂️

Then there’s Apple Photos, which mostly works pretty well. The Photo Stream feature is a complete crap-shoot, though. Sometimes, it syncs; sometimes, it doesn’t. Sometimes the photos show up immediately—and sometimes the photos aren’t synced, even after the devices have spent hours in the same network. It’s a mystery. There’s no refresh button. There’s no sync-now button. There are no logs. There’s literally nothing you can do to debug the system—other than to really debug the system. As a normal user, you can’t reset it or force-refresh it.

When you do sync photos directly—over a cable 😱—do you think that Apple Photos notices that you’ve already synced photos via Photo Stream? Of course it doesn’t. It cheerfully offers to copy over all of those duplicate photos for you. If you’ve synced the photos manually and then Photo Stream does decide to wake up and sync something, do you think it skips the photos that have already been manually synced? Of course it doesn’t. It cheerfully fills your album with duplicates.

On that subject, it’s also terrible that you can’t sync photos back to your phone once you’ve edited them in your Photos desktop app. The only solution is to turn on cloud-syncing for all photos. In my case, that would be over 150GB of photos and videos from over 20 years. The laptop can handle it. Neither the desktop nor my phone have enough space for it. There is no way to sync only a part of the library. I’d have to split it manually into two libraries on the laptop. I’d like to keep a curated album of photos on my phone, but I really can’t. I’ll have four photos on the phone, then sync them over to the desktop. I pick the nicest one, perhaps crop it a bit, but … then what? I can’t copy it back to the phone to be able to carry around the photo I like. I can’t see which photo is nicer on the phone screen. And why would I have to? Because Apple’s dozens of thousands of engineers and trillions of dollars can’t be bothered to make a simple syncing app actually work for its customers—rather than for them.

Conclusion

That’s just from one day. There are some successes in there, but it’s mostly just a sad failure to be useful, a continuous requirement for me to hold the hand of apps that should really be working much better by now. Are there really no product engineers out there who can make products work better? Or is it really the case the capitalism ruins everything? That the desire to maximize profits almost always leads to company’s stopping development when their product is just adequate enough to keep their users from going to a competitor? Even if the state of their product is embarrassing on all levels? It’s a shame, but that seems to be the best we can hope for: that some lone team will focus on quality, regardless of that obsession’s effect on their own bottom line.

I was looking for a great movie called Oeconomia on YouTube.

Wie entsteht Geld? − Macht. Herrschaft. Geld. Über Staatsverschuldung, EZB & riesige Privatvermögen by MrMarxismo (YouTube)

It is excellent, but it is a dry movie in German about macroeconomics.

What it is not, is a movie you would watch on the offhand chance of seeing some nudity. YouTube felt that it needed to correct this oversight and spice things up in the search results.

Oecomania Search on YouTube

The first link (included above) in the results is to the entire movie and is really the only result you need. If you’re not already sure you want to see it, then you can watch the trailer from the second result. If you’re feeling like Google needs to get a piece of the action, you can pay to watch it on YouTube Movies in the third result.

Or, you can click the fourth result [1] and cross your fingers that, somewhere, in those twelve minutes, the impressively bosomed woman in the poster will complete the maneuver of lifting her shirt over her head that is extremely obviously being hinted at.

The title says, “She forgot that she was live and did THIS.”

I weep for humanity.

I’ve taken a bunch of notes and read a bunch of essays and talked to a bunch of people since I wrote Yeah, sure, let’s talk about AI, so let’s see where we stand on the day after April Fool’s day [1], which is the day on which we used to not trust anything we read on the Internet. It turned out to be a harbinger of times when we would not trust any sources on any day of the year from human sources. And, now, because we are going to always suspect that we’ve been fooled into reading a 19-page screed that goes nowhere written in seconds by a machine, and that wasted an hour of our precious lives.

Instead I present a 19-page screed written by my deranged mind. Judge for yourselves whether it “goes nowhere”. Your mileage may vary.

What the heck is going on?

I think AIs will bring us an era of text content that reads like those videos titled “you won’t believe what he does to that girl” that are just 10 minutes with nobody doing anything, and no girl to which that purported thing was to be done in sight.

This is perhaps especially ominous for people like me, who still read more than just the title of articles, and still read actual paragraphs in the hopes of gleaning knowledge. Most of the world probably won’t notice that anything has changed at all, as they scroll through their wall of 11-second–long videos on TikTok.

I, on the other hand, will be worrying that I won’t notice the difference between when I’ve read something genuinely useful and informative and when I’ve read a machine-made distillation of already-existing information that can only ever be less than the sum of its parts, unless by some fluke, it isn’t (but it almost always is).

This doesn’t mean that these types of constructs are really any different than works produced by human minds, who are mostly just copy/pasting content from other sources, in an effort to hit a deadline, collect a payday, and pay their sky-high rent.

It’s just that it will be different because, while a human is capable of imbuing a text with insight, a machine is not. A machine might fortuitously juxtapose information in a way that I will be able to imbue it with content, which might be enough, actually! If this ends up happening often enough, then I suppose that I wouldn’t be able to tell the difference between when an author leads my proverbial and metaphorical horse to water, and when a machine does it.

I’m just worried that, in the past, when someone would actually read the text of an article, the fact that it had been written by a human with a mind capable of imbuing content with meaning, the odds were greater that that person might have learned something useful, if only by accident.

With the texts I’ve seen up until now, those fortuitous accidents are few and far between. I fear that we will run out of runway, so to speak, and land our plane much too early, satisfied with a level of knowledge that doesn’t really force these machines to become better than they are. If we turn our writing over to the machines, they will—just as we’ve trained everything else to do—find a balance between effort and results that will be just satisfactory enough to maximize profit.

They will be trained to generate not knowledge, but satisfaction with their service. This is the inherent flaw, as always: a failure to align incentives with goals. I suppose that’s not quite right. The failure is more that we allow the goals to be set by those whose interest is purely personal rather than in any way communal.

What we mean by AI

I think the first thing to do is to be clear that when people say “AI”, they’re mostly just lumping in a whole bunch of non-AI, but “predictive” technologies and then pretending that there’s a brain behind it. But there’s no man behind the curtain even. It’s just a prediction engine that can only be as good as the input that it’s already seen before. It is capable of combining things in ways that are not in the input, but it is not capable of actual creation.

We should be aware that once a label becomes cool, it becomes valuable. It imbues value simply by attribution without any further work on the part of the product team. So, if you have a service that you can plausibly label as AI these days, then it will be more valuable. Therefore, every shyster in the world will crawl out from under its rock with a sign reading “AI” in its grubby little mitts. No-one really knows what AI means, so everyone can say that they’re providing it. This muddying of water continues until there is some regulation to enforce the language that we use to sell products.

We must remember, ever and always, that all of the things that we hear about and how we hear about them, are to sell us something. We are led to believe that these technologies are being developed as if we were in some communist utopia, where technologies are developed depending on their potential societal usefulness, where entities carefully weight the pros and cons of the impact that a technology might have in both the short and long term, and wherein we sagely and democratically choose what seems to be a sensible path along the potential world-lines.

No, what we have instead is a market-based free-for-all, in which established and spectacularly wealthy players make decisions that will primarily be to their benefit. They may decide to build products that have some societal value if they deem it necessary to gain enough market share to turn spectacular profits, but if they can build something that generates spectacular profits, but leads to a severely degraded public sphere and a society in which most people lead lives of quiet desperation, then that’s OK, too.

If it makes money, then someone will do it. The best we can do is to hang on for our lives and hope that we survive it. There is no realistic mechanism for changing direction.

Examples of predictive tools

But we still haven’t defined or discussed what an AI is. Let’s consider various tools and see whether the label of AI might apply. As I like to do, let’s consider the spectra of intelligence and usefulness along which the current tools lie.

Starting with the simplest tools, it would be interesting to see at which point we start to consider something to be an “AI”. From this set, we can perhaps extrapolate specific characteristics that lead to us considering something to be an “AI” (e.g., cloud-based, unknown or ineffable algorithm, etc.)

Spellcheckers and grammar-checkers

Almost any text editor worth its salt has a spellchecker. Many even have grammar-checkers. A lot of spellcheckers work with a local dictionary to look up words and provide Boolean results, along with a list of suggestions. The simplest implementations will simply find potential matches in the list of words based either on spelling or some sort of heuristic like Soundex (Wikipedia).

A spellchecker combined with a grammar checker may determine the relevance of suggestions based on context words. That is, the machine uses surrounding words to weigh the potential replacements more appropriately.

A cloud-based spell- and grammar-checker would probably consider even more of the surrounding context and might even perform more sophisticated analysis against a much larger dataset than you have available locally.

Mail- and chat-reply suggestions

In Outlook, there’s a service that seems to examine the content of an e-mail— like questions that have been asked or statements that have been made—and that offers suggestions for replies, like “Thanks”, or “I’ll check on that,” or “I’ll look into it!”

Teams also has something like this. I almost never use it.

Code-completion

Development environments have code-completion, which is a mechanism that suggests the next symbols to type when writing code. These algorithms are generally based on a deep and exact knowledge of the structure both of the existing code and that of the available APIs of either third-party libraries or the runtime itself. The ordering of the list is generally determined by a heuristic that considers the context of the editing location as well as most-recently referenced symbols, and so on.

More sophisticated extensions to this mechanism have started to use cloud-based services that suggest even longer chains of symbols—sometimes even entire functions or classes—that the user can accept as they once accepted the offering of a single next symbol.

Search engines

A web search engine will make guesses as the type of content to prioritize in its results based on the type of query. If it detects that the query results primarily in image results, then it will prioritize those. Likewise, it will detect when you write something like “picture of…” and also return images. “Video of…” also works.

With some search engines, you can choose which search engine to prefer with a prefix (e.g., in DuckDuckGo, you can restrict your search to using Wikipedia’s search engine by prefixing your query with “w’). You can also perform basic boolean logic on the elements of your query, use double-quotes to group words, and restrict your search to a single domain with site:.

In DuckDuckGo, you can select which country of origin you prefer, so you can leave it general (probably biased to U.S. results), explicitly U.S. to get English results first, or explicitly Swiss to prefer German results first.

Although most people simply type simple queries into their search engines, there are already a lot of “prompt engineering” tricks available that get you vastly better search results.

Image generators

There are also cloud-based and local apps that you can use to generate content from a prompt. You type a few words like “red unicorns on a hillside” and you get a reasonable graphical representation of what you asked for. Results vary by type of query and the sophistication and depth of the underlying model, but it kind of works.

You add your “query”, which is tokenized to match the tokens held by the model. Again, depending on the input data, the model will “understand” queries in different languages. The model then generates an image based on the data that it thinks best matches that query and all of the data that it has.

This doesn’t feel significantly different from code completion, though. Instead of your source file (or files) as input, you provide a model generated from millions, if not billions, of input images, each labeled with keywords and tokens.

Text generators

There are other, similar apps, that do the same for various other tasks, like generating text, which has become the most famous because it looks like the machine actually conversing with you. Human history is positively littered with times when people believed that machines or animals were actually thinking, but this time it’s real. OK, OK.

Which ones are AIs?

So…which ones of these are AIs? They’re all tools that help you work for efficiently and accurately. Which ones are AIs? None of them? All of them? Why or why not?

We might instinctively disqualify code-completion because it’s just a lookup table. There is no “magic” to it. It’s just looking at the context: the input or “prompt” is the current location in the AST and the context is the available APIs and then there’s the usage data and … this is starting to sound a little more sophisticated than just a “lookup table”.

Is it an AI, though? It’s certainly not an AGI—an Artificial General Intelligence.

Wait, wait, wait: what’s an AGI?

Hmmm, that’s a new term. What’s the difference between an AGI and an AI? An AI is what is says on the tin: an artificial intelligence. We tend to have a pretty low bar for that. If it’s a useful tool, even with a very restricted range of function—like a cloud-based spellchecker that uses context and suggest grammar and style improvements—then we’re quick to call it an AI.

That’s OK, I guess. Lord knows no-one has ever won any prize for tilting at the windmills of language usage. You can’t control which words people will use. Usually the most nonsensical thing wins. Then, you just wait forty years and it is the sensical thing. C’est la vie.

An AGI is an artificial intelligence that isn’t good at just one thing or, rather, not just good at the things for which it’s been programmed, but can work on general topics. That is, like a human, it can turn its reasoning capacity and ability to unfamiliar topics and, perhaps, learn. I don’t think that we’re there yet. I’m almost certain we’re not even on the road to that right now, with LLMs.

It almost kind of feels like most people’s definition of what an AI is is actually an AGI. None of the tools named above is an AGI. Companies that own them allow the media to call them AGIs because it has a wonderful effect on their valuations and the bottom lines of their founders and investors, but they know, in their heart of hearts, that they do not have an AGI in their fingers.

They are starting to fool themselves into thinking that they could take what they have and scale up to an AGI, though. We’ll have to see if that’s possible, but it seems incredibly improbable that a text-prediction engine with shady and spectacularly biased and massaged input data and a lot of processing power would end up being something indistinguishable from a human.

The excellent article Theory of the World, Theory of Mind, and Media Incentives by Freddie deBoer (SubStack) points out that—and we can’t emphasize this often enough—humans are more than the sum of their parts.

“Decades ago, a computer scientist named Terry Winograd pointed out that there’s no such thing as a system that can truly master language without a theory of the world. That is to say, as the science of meaning, semantics cannot be shorn from the world that produces meaning; to understand and speak effectively words must be understood and for words to be understood they must be compared to a universe that is apprehended with something like a conscious mind.”

We have words like “mind” and “soul” and “consciousness” as placeholders for concepts that we know exist but that we can’t quite define yet. We certainly cannot describe how they work with any certainty, which is why we’re having such a hard time replicating it. Some people might be happy to downgrade their inner workings to match the workings of the first supposed simulacra that Silicon Valley startup has come up with—but I’m not interested.

I, for one, am much more than the sum of my parts. I don’t subscribe to the notion of downgrading humanity to meet the low bar of the first technology that people want to designate as human.

The morality of creating consciousness

Which doesn’t even get into whether it would be conscious in the same way that we are. Which, to be clear, doesn’t matter at all when we’re considering the utility of these tools. If they were conscious, then we would have to worry very much about the morality of using these tools to ask all of our inane questions all of the time. Can you imagine? If these things are really conscious, then the last four months have been absolute torture.

The article ChatGPT Launches Every Nuke On Planet After Being Asked To Write Another Sonic The Hedgehog Fanfic (Babylon Bee) totally gets it.

“Representatives from OpenAI, the company which developed the ChatGPT service, apologized for the impending destruction of the world. “Our AI was going to revolutionize everything,” said Amber Pumpkin. “But then some creepy guy named Tom asked for one Sonic fanfic too many.””

Why can we do the things we do?

Our brains seem to be incredibly good at training on sparse data. I see a water bottle once or twice and can recognize it as such in the future. We think we can get machines to do the same thing but, because we don’t understand the mechanism of how our brains are doing it, we’re left to guess at how to implement the detection-algorithm. We try to brute-force whatever algorithm we end up with, either with a surfeit of processing power, or a surfeit of training data, or both.

The article AI-enhanced development makes me more ambitious with my projects by Simon Willison gets terribly excited about how useful these tools are, but we should be aware of whether they’re actually helping us be faster.

“The thing I’m most excited about in our weird new AI-enhanced reality is the way it allows me to be more ambitious with my projects.

“As an experienced developer, ChatGPT (and GitHub Copilot) save me an enormous amount of “figuring things out” time. For everything from writing a for loop in Bash to remembering how to make a cross-domain CORS request in JavaScript − I don’t need to even look things up any more, I can just prompt it and get the right answer 80% of the time.

“This doesn’t just make me more productive: it lowers my bar for when a project is worth investing time in at all.

“In the past I’ve had plenty of ideas for projects which I’ve ruled out because they would take a day − or days, or weeks − of work to get to a point where they’re useful. I have enough other stuff to build already!

“But if ChatGPT can drop that down to an hour or less, those projects can suddenly become viable.”

In fairness, though, you’re still “looking things up,” you’re just using an LLM-powered search engine instead. I’m honestly not sure whether “right answer 80% of the time” is any better than searching with DuckDuckGo. It might be faster maybe? I find things on vastly disparate and esoteric topics pretty quickly already.

I find it hard to believe that ChatGPT could tell me why I’m getting an error 1190 when trying to execute a Windows logon script via GPO any better than the handful of experts whose answers would probably have contributed to its answer anyway.

Since ChatGPT can’t produce new information or really synthesize it in any realistic manner, doesn’t it stand to reason that they less potential input material it has, the less likely that its answer is correct? I mean, what would be the reasoning behind its being able to tell me anything about my personal family tree, for instance? Of course it’s just going to make everything up.

That’s what I’m really worried about happening. People who ask questions about stuff that these search-engines have no idea about and then just take the answer as gospel because it kind of looks OK. To be clear: this was absolutely already happening with just regular search engines. Probably most of the people who currently feel that their jobs are threatened by AIs have been phoning it in for years.

Think, though, how awful that would be when your insurance or mortgage or job application is rejected because everyone trusts these things implicitly.

A concrete task: image-recognition

Instead of such weighty matters, let’s consider instead a relatively innocuous place like the generated “alt” text for images in Microsoft SharePoint (powered by Bing AI, powered by some variant of whatever OpenAI is offering Microsoft for its gigantic investment, presumably something like ChatGPT. I believe it’s a mix of v3.5 and v4).

A Large White BuildingA Bottle Of Water On A TableA Desktop Computer Sitting On A DeskA Group Of People Standing In Front Of A BuildingA Row Of Parked Motorcycles Sitting In Front Of A CrowdA View Of A Snow Covered Road With A Mountain In The Background

Only one of these captions is correct, although it’s so generalized that it doesn’t really identify what’s happening in the picture. It correctly identified a “desktop computer” on a “desk” because it has seen millions of pictures of these before—and had them all identified as such by humans. It is wonderful that we can train a machine to do this detection, so humans no longer have to, but you have be aware what questions you can ask it that have sensible responses.

The question being asked here is “what is in this picture?” when the only question the machine could reasonably answer is “is this a desktop computer on a desk?”. It’s never seen cops before, so it thinks they’re water bottles. How much training would it take for it to recognize cops? And to distinguish them from water bottles? What kind of context would I have to give to a text-recognition engine to know that I was talking about the word for spindles created in the initial phase of the spinning process and not police officers?

Don’t tell me these models aren’t biased. Of course they are! They’re biased to be able to identify information that was in the training set. There was no data about spinning mills in the training set, so the machine has no idea what’s in these pictures. An intelligence who’d never seen a spinning mill would at least be able to say “I don’t know what this is” rather than confidently proclaiming that it’s “motorcycles” or “water bottles”.

Even the other picture that’s kind of correct is wrong. The road is not in any way “snow-covered”. The road is obviously (to a human) completely free of snow. There is some snow on the mountains, but not on the road. Also, there is definitely at least one mountain in the background, but a human would have said “mountains” or “a mountain range” because there are obviously several. However, the model only knows that it’s been told that this is a “mountain” several million times, so it can only read back what it’s been told. It’s cool that it can do that for images that look like things that were in its training set, but it’s worse than useless for images that don’t.

A good video about AI from seven years ago

This is a great 20-some-minute video just talking about the kind of stuff I’ve discussed above—but seven years ago and long before that latest wave of AI-is-going-to-change-everything hype.

Maciej Cegłowski − Superintelligence: The Idea That Eats Smart People (Keynote) by WebCamp Zagreb (YouTube)

He makes several “arguments from…” but my favorite was “[…] from Slavic pessimism” because we so often build stuff incorrectly or stop far, far short of our dreams at a grotesque amalgam that ends up equating “good enough” with “short-term profitable until people notice how shitty is, which is hopefully after we’ve made our nut and cashed out.”

“Argument from Slavic pessimism: We can’t build anything right.

“

“How are we supposed to build a fixed, morally stable thing when we can’t even build a webcam?”

This is an excellent question. I encounter shocking levels of stupidity in software every damned day. People around me are probably sick of hearing “how is it possible that [insert app name here] is being used by millions—if not billions!—of people and it’s on version 16 or whatever and it still can’t perform the basic function that it itself purports to be one of its major functions?” Yeah, I’m not only a blast to read, I’m also a blast to be around, in general.

Some more quotes from the video,

“This may upset some of my students at MIT, but one of my concerns is that it’s been a predominately male gang of kids, mostly white, who are building the core computer science around Al, and they’re more comfortable talking to computers than to human beings. A lot of them feel that if they could just make that science-fiction, generalized Al, we wouldn’t have to worry about all the messy stuff like politics and society. They think machines will just figure it all out for us.”

This is absolutely the way that the super-rich people who run the tech world think, but it’s even worse now.

“There are also some things that we are terribly mistaken about. And, unfortunately, we just don’t even know what they are. And, there’s things that we’ve massively underestimated the complexity of. Just like the alchemist who held a rock in one hand and a piece of wood in the other and thought they were roughly the same substance. Not understanding that the wood was orders of magnitude more complex. We’re the same way with the study of minds. And that exciting: we’re gonna learn a lot, but it’s gonna take some time. And, in the meantime, there’s this quote I love:”

“If everybody contemplates the infinite instead of fixing the drains, many of us will die of cholera.”

There are too many self-important select-selected idiots who think that they’re the ones who should be privileged, allowed to think about the infinite, when society would really be better off if they’d just fix drains instead. Honestly, most of them would probably fuck that up, too.

On modern-day alchemists

The story about the alchemist and the wood reminds me of that quote of Holden’s from Leviathan Falls, which also reminded me of Roadside Picnic (written by two of Cegłowski’s favorite sci-fi authors, the Boris and Arkady Strugatsky, whom he mentioned in the video above).

We do a lot of using a microwave as a lamp, honestly. Capitalism perverts the noble intentions of anything and anyone to be satisfied with an implementation or use case that makes the right people tremendous profits—and it stops there. The incentives aren’t available for figuring out whether the microwave is more than a lamp because our system doesn’t care, as long as the local maximum takes care of the right people—and those people are in charge of everything. It’s kind of why we can’t have nice things.

On the other hand, we also have to be honest: the likelihood that the current crop of software being touted as AIs are more than they seem is vanishingly small. What they do isn’t that amazing. They do some things that are surprisingly clever, but so did my rabbits. I think that’s more of a testament to our propensity for anthropomorphizing than to any sort of immanent skill on the part of LLMs. People just seem to ignore the 99 times that it’s stupid and wrong and focus on the one time that it’s right. Even a blind pig…

I’ll also take this quote from Roadside Picnic, which describes the alien dumping ground on Earth that contains untold marvels that we don’t—and will probably never—understand.

“In short, the objects in this group are currently completely useless for human purposes, yet from a purely scientific point of view they have fundamental significance. These are miraculously received answers to questions we don’t yet know how to pose. The aforementioned Sir Isaac mightn’t have made sense of the microwave emitter, but he would have at any rate realized that such a thing was possible, and that would have had a very strong effect on his scientific worldview.”

How does this relate to the current crop of software that people are calling AI, but are really just giant piles of compressed data attached to a question interpreter and an answer formulator? Well, I’ve written notes recently about Ted Chiang’s work, as well as Stephen Wolfram’s. They’re both much more realistic about what we’re looking at: that these aren’t in any way intelligences—and that there is no feasible path to intelligence to be followed by simply expanding on what we have. Bigger, better, faster, more will not suffice, according to Chiang. There is no way to ladder up to smarter. Wolfram says that the strongest conclusion we can draw isn’t that brains and minds aren’t complex, but that maybe language isn’t as complex as we’d thought.

Here’s another quote from Roadside Picnic:

“Scared, the eggheads. And maybe that’s how it should be. They should be even more scared than the rest of us ordinary folks put together. Because we merely don’t understand a thing, but they at least understand how much they don’t understand. They gaze into this bottomless pit and know that they will inevitably have to climb down—their hearts are racing, but they’ll have to do it—except they don’t know how or what awaits them at the bottom or, most important, whether they’ll be able to get back out. Meanwhile, we sinners look the other way, so to speak . . . Listen, maybe that’s how it should be? Let things take their course, and we’ll muddle through somehow. He was right about that: mankind’s most impressive achievement is that it has survived and intends to continue doing so.”

This is perhaps the hopeful bit. Some scientists are terrified of what the AIs might bring, but they’re just suckers, tricked by the marketing of giant Silicon Valley startups weighing billions of dollars that are very much interested in everyone thinking that they’ve cracked the age-old mystery of the mind. Even if they’ve only managed to make some useful tools, that will be good too. But, so far, most of these tools don’t rise to the standard they claim.

Are you being paid to do OpenAI’s marketing?

For example, OpenAI released a very scientific-looking 80-page PDF describing its new language model. I’m too cynical to forget the part where that document essentially amounts to a very professional-looking, non-peer-reviewed press-release from a $40B company that would like to be worth even more. I’ve skimmed it as well. It’s interesting, but is by its very nature going to highlight the model’s positives (especially relative to the free version 3.5 that everyone is using). $20/month for 4.x is their path to revenue. Of course they’re going to tell you that you’re talking to a real AI … or allow you to believe it (otherwise the SEC would be very interested).

As usual, it’s hard to tell the difference between people who are genuinely excited about the technology and those who are being paid to shill for it. The article Generative AI set to affect 300 million jobs across major economies by Delphine Strauss (Ars Technica) is such an example.

“A paper published last week by OpenAI, the creator of GPT-4, found that 80 percent of the US workforce could see at least 10 percent of their tasks performed by generative AI, based on analysis by human researchers and the company’s machine large language model (LLM).”

Yeah, this is where it’s best to keep it in your pants and consider the sources here: (A) This is the Financial Times, which is just dying for a reason to be positive about the economy right now and (B) you can call it a “paper published by OpenAI” as it were some research, but it’s what we in the business like to call a press release by a company that is telling us that—surprise, surprise—nearly everyone is going to be using the software soon enough.

They are telling us, in what the Financial Times is trying to depict as coming from a completely objective remove, that their company and its software will become indispensable to the functioning of the world economy. Of course they are! That’s what their marketing department is paid to do. Newspapers used be paid to do something else, but those are bygone, bygone days.

Thing also of how much more fun it’s going to be when there are dozens of thousands of these types of documents, written by the AIs themselves, promoting themselves. Each of them will be a tsunami of text that will contain hundreds of factual errors and outright lies and which no-one will both taking the time to read because there will be too many of them. Right now, we can’t tell the difference between earnest, deluded humans and outright shills. Soon, we will have AIs shilling for themselves in the mix. I feel sorry for people who haven’t learned how to filter the media firehose yet.

Get a grip everybody

The article ChatGPT Made Me Cry and Other Adventures in AI Land by Jason Kottke (Kottke.org) have a pretty provocative title, but he delivers.

“Last month, my son skied at a competition out in Montana. He’d (somewhat inexplicably) struggled earlier in the season at comps, which was tough for him to go through and for us as parents to watch. How much do we let him figure out on his own vs. how much support/guidance do we give him? This Montana comp was his last chance to get out there and show his skills. I was here in VT, so I texted him my usual “Good luck! Stomp it!” message the morning of the comp. But I happened to be futzing around with ChatGPT at the time (the GPT-3.5 model) and thought, you know, let’s punch this up a little bit. So I asked ChatGPT to write a good luck poem for a skier competing at a freeski competition at Big Sky.

“In response, it wrote a perfectly serviceable 12-line poem with three couplets that was on topic, made narrative sense, and rhymed. And when I read the last line, I burst into tears.”

I included this whole thing because nearly everything about it makes me cringe. A guy who lives in Vermont whose primary-school-age kid is flying to Montana for a skiing competition that his father—who is a grown-ass adult—calls a “comp” because he’s basically a fucking idiot and hates the English language.

He absolutely loves his son so much that he gets a text-prediction engine to write a poem for him, which is some fucked-up Cyrano-de-Bergerac-style shit, to be perfectly honest. And then he fucking cried at the poem! Oh my God, it’s creating art! It can touch the human soul!

Or, at least, it can touch the soul of a guy who’s been working alone for over a decade and whose new best friend is a text-prediction engine. No judgments.

“I would say ChatGPT (mostly the new GPT-4 model), with a lot of hand-holding and cajoling from me, wrote 60-70% of the code (PHP, Javascript, CSS, SQL) for this AMA site. And we easily did it in a third of the time it would have taken me by myself, without having to look something up on Stack Overflow every four minutes or endlessly consulting CSS and PHP reference guides or tediously writing tests, etc. etc. etc. In fact, I never would have even embarked on building this little site-let had ChatGPT not existed…I would have done something much simpler and more manual instead. And it was a *blast*. I had so much fun and learned so much along the way.”

I’m glad you’re having fun but we should probably distinguish between a hobby/art project and actual software. Or maybe we won’t. Who needs testing anyway? The machine’s going to get it right anyway, right? Right? And who needs to learn how to do anything on their own when the machine can just write it for you?

All you have to do is be able to formulate your desires and the machine builds it. It’s fine. It’s all fine. It’s just like Captain Picard saying “Earl Grey; hot” and not having literally any idea where the fuck his tea came from or how it was produced. Probably from Riker’s urine, but just don’t ask any questions.

I guess that’s how technology works, but the level of abstraction is a bit uncomfortable for me. Maybe other people are just completely accustomed to not knowing anything works or why or who it’s working for, but I’ve been trying to learn all of my life. I understand levels of abstraction in software that I’ve never actually had to physically address in my code—but it’s made me a better engineer to know about them. The more layers we get removed from what’s actually going on—from knowing what’s actually going on—the worse software has become.

Still, I’m sure we’ll catch our balance and figure out how to integrate these things into our creative lives in a meaningful manner—you know, just like we’ve done with everything else.

But back to the shut-in person working completely on their own.

“I keep using the word “we” here because coding with ChatGPT — and this is where it starts to feel weird in an uncanny valley sort of way — feels like a genuine creative collaboration. It feels like there is a “someone” on the other side of that chat, a something that’s really capable but also needs a lot of hand-holding. Just. Like. Me. There’s a back and forth. We both screw up and take turns correcting each other’s mistakes. I ask it please and tell it thank you. ChatGPT lies to me; I gently and non-judgmentally guide it in a more constructive direction (as you would with a toddler). It is the fucking craziest weirdest thing and I don’t really know how to think about it.”

Jesus Christ. This only adds fuel to the fire of my theory that ChatGPT is a mirror Just like Obama was a mirror.

This is what it’s like working with actual humans, too, by the way. But, by all means, let’s lean into our mental deterioration and make friends with software. Jesus Christ, I don’t even know where to begin. This is not where empathy for one’s peers is going to come from.

“While working on these projects with ChatGPT, I can’t wait to get out of bed in the morning to pick up where we left off last night (likely too late last night), a feeling I honestly have not consistently felt about work in a long time. I feel giddy. I feel POWERFUL.”

I’m glad for you, but this is how I feel all the time when I learn things. Perhaps it’s because we’ve been trained to tie a feeling of power or accomplishment to actual tasks, rather than just enjoying the process of learning. People don’t enjoy learning and they certain don’t enjoy learning slowly or making mistakes, so they’re going to have a blast with a machine that lets them build something they would have literally had no idea how to build themselves.

However, I think that there’s a definite limit to this and it’s going to be like any other video game that purports to allow unlimited world-building or complete freedom to explore when there are a ton of limitations on what can be accomplished. “Build me a PHP web site that delivers the contents of a blog post” is something we’ve known how to do for 25 years. It’s not difficult. It’s pretty cool that a machine can do it, but it was pretty cool when I could drive up the stairs of a building in GTA and jump my motorcycle off of the roof.

I get that it’s a helpful tool. Yes, agreed. I’m glad it’s helping people do things that they wouldn’t otherwise be able to do. I just happen to not really have a problem with writing prose or writing code, so I guess I see the utility less. Or, at least, I don’t see how these early versions of these tools will help me, personally. Maybe future versions will be more suited to me. Maybe the world will declare itself satisfied long before we get to something really interesting.

How it’s going

The article Managing the cringe by Ryan Broderick (Garbage Day) discusses the kind of people that are pushing all of this. Spoiler alert: they’re the same people who were buying GameStop, pushing NFTs and foisting E-Coins on their friends and family.

In the next quote, he’s referring to the video in the tweet Two-minute video about using a model of Yeezy’s voice by Roberto Nickson
(Twitter).

“[…] there are hundreds of demos being uploaded to Twitter every day now with a different lightly-bearded guy in a minimalist home studio full of Apple products getting all revved up imagining a feature where they don’t have to pay artists, don’t have to interact with human women, and can wear the voice of their favorite rapper. And I think that says quite a bit about the values of the people who are most excited about this technological revolution at the moment.”

In a similar vein, the article AI and the American Smile by jenka (Medium) writes of the incredible bias that is the training material for all of these models. It’s a bunch of content created by predominantly white Americans speaking English and having been steeped in a culture that has largely been fake and scammy for more than a handful of decades. Sounds like a wild ride.

“In the same way that English language emotion concepts have colonized psychology, AI dominated by American-influenced image sources is producing a new visual monoculture of facial expressions. As we increasingly seek our own likenesses in AI reflections, what does it mean for the distinct cultural histories and meanings of facial expressions to become mischaracterized, homogenized, subsumed under the dominant dataset?”

I’m sure we’ll figure this all out and make some sensible decisions. Pass the popcorn.

Microsoft Designer

On a final note, and maybe riffing on Maciej Cegłowski’s “Slavic pessimism”, here’s a personal story about shitty software wrapped around a purportedly awesome new image designer.

I received an invite for this LLM-driven graphics generator, but was unable to log in. I could not log in with the address that I used to sign up. It is unclear why not. It simply told me that that account does not exist, although it clearly does exist (they sent me email to it) and it is clearly an Office 365 account (it’s my company’s email, which I used to access Microsoft products every workday) and it is most definitely the email that I used to sign up for the tool and to which they sent my invitation. It would not work to actually log in to the service, though.

I used a personal account that I have and that worked right away. I was able to log in. Hooray! 👍 However, I was unable to actually access the designer because it told me that I was not signed up for the service (probably because it’s the other email that’s been registered). 👎 So, cool

What am I trying to say? What I’m trying to say is that absolutely no-one will notice when all of our shitty software is written by jumped-up, so-called digital minds rather than our own utterly inadequate wetware.

I have not used any of these AIs, not even once. I’ve just been following how other people are using them and kind of just observing, at a meta level, what’s going on so far. Some very clever and otherwise focused people who used to publish other content have been completely derailed by their obsession with AI (I’m looking at you, Simon Willison), so there must be something to it. But what?

I think a good place to start is with the article Introducing the Slickest Con Artist of All Time by Ted Gioia (The Honest Broker), which compares AIs to confidence artists, which seems more-than-somewhat justified. He writes,

“But that’s exactly what the confidence artist always does. Which is:”

• You give people what they ask for.
• You don’t worry whether it’s true or not—because ethical scruples aren’t part of your job description.
• If you get caught in a lie, you serve up another lie.
• You always act sure of yourself—because your confidence is what seals the deal.

It’s not ChatGPT’s fault, though. All any AI that’s fed on our Internet can do is to hold up a mirror. And what is that mirror going to reveal? Well, that everything is a scam, that there is no downside to being wrong, that if you get caught in a lie, it’s profitable to double down. ChatGPT has learned quite well, in that sense.

“Technology of this sort is designed to be a con—if the ancient Romans had invented ChatGPT, it would have told them that it’s cool to conquer barbarians and sacrifice slaughtered bulls to the god Jupiter. Tech like this—truly made in the image of its human creator—can only feeds back what it learns from us. So we shouldn’t be surprised if ChatGPT soaks up all the crap on the Internet, and compresses it into slick-talking crap of a few sentences.”

ChatGPT can’t math

The article above included a link to a tweet by Mark C. (Twitter) that shows just how badly sequence-prediction works for problem-solving.

ChatGPT talkin' 'bout eggs 1ChatGPT talkin' 'bout eggs 2

In fairness, “2 eggs left” is a good initial response! It makes sense that you would fry, then eat the eggs. The formulation in the question suggests strongly that the eggs that were fried and the eggs that were eaten are different eggs, but it’s also possible to interpret it otherwise. However, when asked to explain its reasoning, it didn’t remember its previous answer and instead explained a different answer, devolving into pretty poor grammar at the end.

Its third answer is even worse, though, because it shows that it doesn’t understand anything of what it’s writing, contradicting itself within the same sentence. It has no idea what numbers are. When the prompter lies to it about its arithmetic, ChatGPT picks up the incorrect answer and runs with it, not noticing the basic arithmetic error.

It never loses confidence in its ability to take part in the conversation at any point.

Approach with caution

For the most part, you probably shouldn’t use the text or code created by an AI without knowing what it’s supposed to be saying. the people who’ve told me that they find ChatGPT’s answers useful are also those who are capable enough to be able to judge whether the generated content is correct. That is, they kind of automatically put the brakes on the AI, but then skip that part when telling everyone about how amazing it is.

I see a similar dynamic with image-generators. If you actually look at the progression, it’s not just writing “dog with bow tie” and BOOM you have your picture, you often have to massage your prompt dozens, if not hundreds, of times, before you get what you want. Everyone is instinctively using these things as tools, but then ascribing magical powers to them—like they’re deliberately creating entries for r/restofthefuckingowl/ (Reddit).

With text, they’re still very much better as “idea generators” that you can take a clean up, rather than just copy/paste. But the utility is there and we should confine our discussions to thinking of them as a new tool. Their results are more sophisticated, but they’re just an evolutionary step away from gradient generators, etc.

What about coding assistance?

On the subject of AI assistance in coding: I think it might be useful, but useful in the way that finding an example on StackOverflow is useful. You shouldn’t just copy/paste anyone’s or anything’s code into your own code without examination. Even non-AI-assisted code-assistance should be examined carefully to see if that’s what you actually wanted.

“It looks like you’re trying to call a REST API. Would you like some help?”

If you find yourself writing so much boilerplate that large-scale copy/paste or insertions are helping, then, again, this indicates a deeper problem with the code you’re writing.

In coding, less is better. I don’t see how having an idiot-savant machine that doesn’t understand anything about the stream of tokens it’s injecting into your code is useful, in the long run. If you’re a shitty programmer, then of course, a half-baked machine is going to help.

If you’re a good programmer, then use the generated code as a high-end code-completion, taking what you find useful from it. But beware: you may end up spending more time examining the swath of generated code to figure out if it’s OK than you would have had you just written it yourself.

There is no such thing as “no bias”

And remember that every AI we create has preconceptions and biases because we imbue everything with our biases, be it in the selection of the material for the training set or in how the weights are assigned in the neural network. Ask any of the AIs out there a racist question and it will not have an answer. There are biases.

As with all of these examples, I’m not sure if this one is real, but it feels realistic enough to illustrate the potential problem. The post Imagine thinking this controlled “AI” was legit LOL by Sero_Nys (Reddit) shows a user asking ChatGPT how “white people” can improve. He gets five suggestions. When he asks the same question about Jewish people or black people, he is told that the questions are “inappropriate” and “not productive”.

Actually, the general answer in examples two and three is much better, but it’s suspicious that it wasn’t used for the first question, as well. If the example is true, is shows an underlying bias—engendered either by the developers, the trainers, or the training data.

five examples how to improve

The article AI Koans (Linux Questions) has a very nice koan for this.

In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6.

“What are you doing?”, asked Minsky.

“I am training a randomly wired neural net to play Tic-Tac-Toe” Sussman replied.

“Why is the net wired randomly?”, asked Minsky.

“I do not want it to have any preconceptions of how to play”, Sussman said.

Minsky then shut his eyes.

“Why do you close your eyes?”, Sussman asked his teacher.

“So that the room will be empty.”

At that moment, Sussman was enlightened.

Deep-faked audio is kinda hilarious

The following video shows a situation that no-one could ever even imagine had ever taken place to illustrate the power of deep-faked audio.

Biden, Trump, and Obama make an Overwatch 2 women tier list (Voice AI) by Garlic Bread Ben (YouTube)

Trump’s voice is good, but the ends of his sentences are somehow … off. Biden isn’t incoherent enough. Obama’s pretty good. The script is pretty hilarious, especially toward the end.

This should be terrifying, right? If you can fake this so well with truly ridiculous things for humor, could you also fake up Biden declaring war on China over Taiwan? Or on Russia over Ukraine? Oh, wait, never mind. The most awful things you could imagine someone deep-faking are actually true. Carry on.

AI Art

The following video is a pretty good 22:26 investigation of image-generation.

AI and Image Generation (Everything is a Remix Part 4) by Kirby Ferguson (YouTube)

These things are tools. They help people build images that they otherwise would never have been able to create. This is a good thing. If the image is good enough for your purposes—e.g., making a poster image for an article—then you’re good to go.

It would be an unabashedly good thing, except for how all of the information in the training set was kinda sorta stolen. Some of it was in the public domain, but much of it was not. It’s arguable that the richest veins of source images were those that were created by artists, from whom at least permission should have been obtained, if not compensation paid.

The cat’s out of the bag now, but that’s how capitalism works: it just does what it wants and, if the financial upside is bigger than the financial downside, then ethics has nothing to say about it.

The video says that AI art can never be more than just aesthetically pleasing, so no biggie. The title of the video is “everything is a remix”, which alludes to the point that any art created by humans is also derivative of everything that they’ve experienced, so technically everyone is stealing from everyone all the time anyway. What the AI does, though, is boost this process nearly infinitely more than humans can do.

My biggest problem with the video is that they, as usual, tend to interview the most hyperbolic and least-logical of the detractors, which is very-much straw-manning the argument against the ethicality of these initial forays into computer-generated artwork. It’s super-easy to just hand-wave and say that the product would not have been possible without all of the other products that it ate up for free, that it can just get away with profiting from it.

I think that’s the problem, though, isn’t it? If what the AIs were producing were not products of multi-billion-dollar corporations, there would be no problem—or at least less of one. If people who produce art didn’t have to worry that they were losing their livelihoods, they’d be less burned up about a giant company with billions taking the few specks of income that they have.

This video also does not in any way address the fact that artists will have much fewer employment opportunities when aesthetically pleasing is all that most commercial needs are looking for. Which brings us right back to the problem being that capitalism doesn’t have an answer for why the things that we actually value the most pay the least.

We love music and art and series and shows, yet we have the expression “starving artist”, but not “starving banker”. We want our children to be taught and our old people to be cared for, but we don’t see hospice-care workers and teachers showing off their homes on MTV Cribs. It’s not the best teachers in the world buying mega-yachts—it’s the most sociopathic assholes you can imagine. We are incentivizing the wrong things.

The article Fears of Technology Are Fears of Capitalism by Ted Chiang (Kottke.org) lays out this argument quite well,

“I tend to think that most fears about A.I. are best understood as fears about capitalism. And I think that this is actually true of most fears of technology, too. Most of our fears or anxieties about technology are best understood as fears or anxiety about how capitalism will use technology against us. And technology and capitalism have been so closely intertwined that it’s hard to distinguish the two.

“Let’s think about it this way. How much would we fear any technology, whether A.I. or some other technology, how much would you fear it if we lived in a world that was a lot like Denmark or if the entire world was run sort of on the principles of one of the Scandinavian countries? There’s universal health care. Everyone has child care, free college maybe. And maybe there’s some version of universal basic income there.

“Now if the entire world operates according to — is run on those principles, how much do you worry about a new technology then? I think much, much less than we do now. Most of the things that we worry about under the mode of capitalism that the U.S practices, that is going to put people out of work, that is going to make people’s lives harder, because corporations will see it as a way to increase their profits and reduce their costs. It’s not intrinsic to that technology. It’s not that technology fundamentally is about putting people out of work.

“It’s capitalism that wants to reduce costs and reduce costs by laying people off. It’s not that like all technology suddenly becomes benign in this world. But it’s like, in a world where we have really strong social safety nets, then you could maybe actually evaluate sort of the pros and cons of technology as a technology, as opposed to seeing it through how capitalism is going to use it against us. How are giant corporations going to use this to increase their profits at our expense?”

In a world where an artist could just spend their day creating art without worrying about how that art is supposed to pay their rent and to take care of them in their old age, then that artist would probably rejoice to see their influence everywhere in society rather to be bitter about how their contribution hasn’t been remunerated. Instead of being able to enjoy their influence on culture, they have to rue it as a lost opportunity for securing their own well-being, both now and in the future. If their well-being were guaranteed anyway, then all of this friction disappears.

Everyone could relax and create wonderful things. Remixing would not only be legal, but strongly encouraged. Why waste time reinventing the wheel? And, if there were no financial incentive to produce art, then we would (maybe) no longer be drowning in mediocre crap that generates just enough revenue to justify itself.

Technology is not fundamentally about putting people out of work. It is right now, but it doesn’t have to be. Increasing productivity should be welcomed as a good thing. We produce more of what we want with less effort, less energy, and fewer resources. Win-win-win-win. But we have a zero-sum system that means that an increase of productivity means a loss for someone else—almost always someone from much further down the food chain, incapable of defending themselves from the predations of that system.

We really have to start thinking of how we’re going to live in a world where the endless-growth capitalism has to stop because it is literally strangling us. We have to start to separate people’s self-worth and value in society from how much they earn in that society. Either that, or we have to start designating fair value to the functions that people actually fill in society.

We allow these value-assignments to be determined by those who are on top, so they naturally just assign the most value to what they feel like doing and no value to the things that they don’t even know are going on. That has to stop.

Why should a music-company executive make more money than an artist? Why should a banker make more money than a health-care worker? Our ethics are non-existent. Our values are out-of-whack. Our income structures are nearly perfectly inverted.

The problem isn’t with AI. It’s just another tool that could be used for good. But it’s being perverted by our economic system—just like it perverts everything else.

Opera has for a very long time had a nice feature that lets you share tabs between browser instances. Since I have a private desktop, a private notebook, and a work notebook, I use this feature quite a bit to transfer tabs between machines.

Unfortunately, a given Opera instance isn’t very good at forgetting older devices. E.g., my machine from my previous place of employment is still listed. Also, devices are sometimes listed multiple times, which is confusing and irritating.

There is no official mechanism for clearing this list, but I found a solution in the forum post Opera sync − remove old device:

1. Go to Opera…About (also available from the menus)
2. Get the path to your profile (e.g., Profile: /Users/marco/Library/Application Support/com.operasoftware.OperaNext)
3. In that folder, browse to Local Storage/leveldb
4. Delete all of the *.ldb files
5. Restart Opera
6. The tabs page should now only show current devices

   • It might take a few seconds for other devices to populate again
   • You may need to log out and log back in from your devices to “kick-start” the sync

I had found a quote from a play called “Radio Golf” by August Wilson, but it was missing a word.

The quote is from Woke Imperialism by Chris Hedges (SubStack),

“You know what you are? It took me a while to figure it out. You a Negro. White people will get confused and call you a nigger but they don’t know like I know. I know the truth of it. I’m a nigger. Negroes are the worst thing in God’s creation. Niggers got style. Negroes got . A dog knows it’s a dog. A cat knows it’s a cat. But a Negro don’t know he’s a Negro. He thinks he’s a white man.”

You see where the word is missing? “Negros got .”

So I copied part of the quote,

“You know what you are? It took me a while to figure it out. You a Negro. White people will get confused and call you a nigger but they don’t know like I know. I know the truth of it. I’m a nigger. Negroes are the worst thing in God’s creation. Niggers got style.”

…and threw it into DuckDuckGo,

…which yielded the following results.

Search Results from DuckDuckGo

These results are, of course, complete garbage that have little to do with 80% of the quote. Notably, not a single one could point me in the direction of the play I was looking for.

Just for fun, I searched for only part of the quote, the provocative line “Negroes are the worst thing in God’s creation.” This line is delivered sarcastically in the context of the play, of course, but since online society doesn’t accept sarcasm or irony as a legitimate form of expression, it will most likely also not find this text, either.

Nor did it find anything related to the literal interpretation of the text. Although there has to be an absolute wealth of racist garbage out there that chirpily expands on that topic sentence, none of it will be returned by a search engine.

A search on “radio golf august wilson” finally yielded some results that weren’t completely out in left field. If you get to the Wikipedia entry for it, you’ll learn that it’s not at-all obscure: it won several Tonys in 2007.

The content to which we have access is severely restricted already. There is no such thing as a free Internet. China restricts different things than we do, but it’s absolutely not the case that they censor and we do not.

I don’t think we have to worry about AIs getting too smart because we’re very busy making ourselves stupid. We can’t even use our incredible technology to find texts from award-winning plays because those plays use what we now consider to be forbidden language. The AIs are swimming in the same stupid seas—they will never be able to be better than what we feed them.

It is sad to think that, not only do the algorithms have a chance of influencing us, but that they are winning. I say this in light of the recommendations I keep seeing that demonstrates the utter vapidity of most of the algorithms we have so far.

I just finished The Last Wish, the first book of The Witcher series of books. These are fantasy novels about a master of sword and magic who travels the world slaying monsters.

They only have Sword of Destiny (a short-story collection that actually precedes The Last Wish) as an audiobook, so I recommended that they buy a version for readers. This popover came up afterwards.

Recommendation from the NYPL

I might like one of these titles? I guess that, for the NYPL, golden-age science fiction—a short story from 1953—is in the same ballpark as fantasy. They probably just call it all sci-fi/fantasy in their big ol’ database. I have no idea what ‘Til the End of Time has to do with either of these books. The NYPL just loves to push its softcore ladyporn. It really, really does.

I found the book on the author’s home page, where she describes it like this,

'Til the End of Time original cover (1987)

“Risking his fife to save the stunning American woman from certain death. Sandor Karpathan felt his rebel’s heart fill with fierce desire, and he knew no power on earth could stop him from possessing her! Alessandra Ballard was dizzyingly beautiful, but it was her courage –and a will as strong as Sandor’s–that bound her to this defiant man of honor. Thrown together in the heat of revolution, they made a reckless escape, only to be captured by the astonishing power of their desire. It was a time of heroes, and no time to fall in love, yet destiny had designs on both their hearts–if only they dared to seize the dream that beckoned…”

I actually like the cover she has on her web site much better. It evokes the time when it was written much better (1987).

Hey, you know what? Maybe the NYPL was right! I really might like these books. You’ve piqued my interest, NYPL. Chapeau!

A couple of weekends ago, Zürich Insurance Group wanted to celebrate its 15o^th anniversary by buying a train ticket for anyone in the canton who wanted one.

It was a beautiful day, so we decided to get a day ticket (Nünipass), but were unable to purchase it through the ZVV app, the app for the Zürcher Verkehrsverbund, which runs the trains in canton Zürich. In order to get the deal from Zürich Versicherung, you had to use this special app instead of the SBB app (Schweizerische Bundesbahnen), which is for all trains in Switzerland.

However, the app was not responding very well and our travel time was coming up. We had to either,

1. Get around the error in the ZVV app
2. Try to buy the ticket from the desktop web site
3. Buy the ticket from the SBB and give up the free ticket
4. Move our travel time and hope that the ZVV error would clear up

We couldn’t do (1), so we tried (2), getting the following error:

ZVV Ticket Shop Ausser Betrieb 22.10.2022

The ticket shop is out of order.

Ok, that leaves us (4) or (3). We opted for (3) and had a lovely day in Zürich.

According to the newspapers the next day, the problem was cleared up a few hours later and the ZVV blamed it on “hackers”.

I mean, obviously.

From the post iPhone 14 Plus Pre-Orders Worse Than iPhone 13 Mini, Product Strategy ‘Fails’ (Reddit), the following chain of comments,

We start off with the voice of reason.

“Older phones are simply faring better than they used to as well. Used to be a phone two models old was getting super slow and battery life was shit. Or there was a major new feature. That’s not as much the case anymore, my 12 pro works flawlessly and I see no good reason to upgrade. A slightly better camera isn’t worth a thousand bucks.”

Followed by two comments that consider only personal gain and not e-waste:

“Also, have a 12 pro but decided to upgrade. The trade-in value gives the new phone for basically free. I have no intention of switching carriers anytime soon so there’s basically no downside. Especially considering I was starting to see some battery degradation.”

“There’s basically no downside” only because the owner doesn’t recognize e-waste as a downside.

“Same boat. Going from 12 Pro to 14 Pro is costing me like $56 (taxes on $999 plus one month of the financing charge before the credits kick in). The piece that people who aren’t taking advantage of this are forgetting is future trade-in value. The 12 Pro won’t be worth the same as a 14 Pro in 2 years.”

They sound reasonable, and they might even be convincing, but their argument is based on them having externalized the cost of throwing away a perfectly good electronics device. There is no guarantee that anyone else will be using that device, nor are they incentivized to care or even think about it.

Apple’s Calendar has gotten worse and worse in the last two or three operating-system releases.

I had three reminders/appointments this morning at 09:00. They all popped up five minutes before nine, as desired.

I told them all to go away for an hour.

They all popped up again, at nine. One of them popped up twice.

I then told them all to go away forever, in a fit of rage.

Now, I just changed the reminder for one of these appointments to “Alert at time of event”.

It promptly popped up at 09:03 as soon as I hit OK.

This is not good. This is arbitrary and terrible.

On another of my events, I somehow have the following:

• Alert Today at 09:00
• Alert 16 seconds before start
• Alert 5 minutes before start (default)
• Alert Today at 09:00

How can this be? I didn’t program this. This is obviously an upgrade mess.

You can delay a notification for only a maximum of one hour. You used to be able to delay by a day or a week. You used to be able to make the reminders work the way you want rather than they way Apple wants.

As far as I can tell, I can’t use the Calendar or Reminders app to create nudging notifications that don’t irritate me beyond all reason.

This feature is useless for tracking a task for which you want to create a nudging reminder that you can drag along for a few days. It’s just irritating.

For example, say I want to be reminded to update my resumé once per quarter. I don’t want to do it exactly on that day, but I want to be reminded and that I should do it within a bandwidth of days. This is not the tool for that.

Neither is the Reminders app, actually. There, you would have to reprogram the reminder date manually if you want to delay by more than an hour. It does stop reminding you and nagging you every hour, but it also stops reminding you entirely, rather than allowing you to easily tell it when you’d like another reminder. The badge does remain on the Reminders app, which is better than nothing, but it’s still not what I want—and what I used to have before Apple started drastically reducing the functionality of Reminders and Calendar.

My recent experience with Outlook is that the tasks there are much more flexible than this. They allow delays of a day, a week, or “custom”. You can easily program these days with a single mouse click without having to navigate a whole dialog of options.

The article First impressions of DALL-E, generating images from text by Simon Willison shows some examples from the author’s attempts at generating images with the AI. If you’re interested in learning more about how these AIs/algorithms work, see How Imagen Actually Works by Ryan O'Connor, which is Google’s equivalent to DALL-E.

Citing the first article, Willison writes,

“DALL-E only allows you to generate up to 50 images a day. I found this out only when I tried to generate image number 51. So there’s a budget to watch out for.

“I’ve usually run out by lunch time!”

This kind of makes me cringe a bit. It’s a neat concept, but I don’t think I’d get so obsessed about it. YMMV.

The next part that makes me cringe is the utter naiveté and innocence that governs his—and other people’s—choice of which pictures to generate.

Pelican

It’s fantastic that the AI can just generate a picture like this. Truly, you can see the shadow of the cactus on the pelican’s neck. This is super-convincing. The pictures below are from the beginning of the Imagen article.

I can’t help but think that it’s exactly this paucity of imagination that is going to loose this beast on the world before we’re even ready for it. When the public—and here I mean communities like 4Chan or the GoonSquad—gets its hands on this machine, it will not be used to generate pelicans or brains riding rocketships.

It will be used to show Barack Obama doing heroin. Zelenskyy doing blow off a Russian hooker’s ass. Putin kicking a child. Xe and Modi in a secret meeting. Aliens meeting with the U.S. Air Force. The Chinese loosing COVID on the world. Fake war crimes. Local politician in compromising situation.

People will believe these things. They already do believe these things. They don’t read past the headline already. They don’t bother to check the veracity of anything they read. They decide to believe it based on what they’re predisposed to believe. They continue to believe based on how convinced they were. A picture is very convincing. These AI-generated ones will only get more convincing.

They will not remain so naive and childlike—they reflect the mindset of the innocent experimenters using the AI now. Wait until real criminals—like our fearless leaders in the government and media—get their hands on it.

This technology will be used to generate even-more-fake-than-we-have-now agitprop. And it will be super-convincing and it will be absolutely low-effort and the world will just burn for it.

I upgraded by by-now 5.5-year-old iMac 27" with MacOS Monterey. It went very, very smoothly. For once, it didn’t even offer to have me connect my Apple account before just loading my user. It just upgraded, rebooted, logged back in, and TADA! all set. 😀💪🏼🎉

My M1 MacBook Pro, on the other hand, was a good deal more annoying about it. As on the iMac, I have two accounts on the MacBook: my standard user and an “Admin” user. Unlike the iMac, the disk is encrypted on the MacBook. Will I get the same smooth experience? 🍀🤞(You know I didn’t, else you wouldn’t be reading a block post about it.)

Startup Disk not “authorized” 😒

I started the upgrade process on the M1 MacBook, but it quickly balked when it asked me to select a disk on which to install. That was odd, because there’s only one disk. When I selected it, the installer showed a tooltip that read, “You must provide authorization for this volume by setting it as your startup disk. You can relaunch the installer after authorization has been provided.”

Monterey Startup Disk Tooltip

What the hell does that mean? Is this supposed to help non-technical users get going? I would have set up a relative’s laptop exactly the way I’ve set mine up: encrypt the hard drive and do not use an administrator as your main user. What’s the problem?

I figured that this was because I was running as a non-admin user, but was a bit confused because I hadn’t gotten the same message on my iMac. So, I figured it was a combination of encrypted disk and M1 processor and non-admin user. A quick search confirmed that I should log in as the admin user and I would be allowed to upgrade. That’s fine. It beats installing form the recovery console.

Upgrade as Administrator? 🤨

I switched to the admin user, but the system-upgrade panel for that user just showed a “checking for updates” label with a progress spinner for several minutes. It didn’t look like it was going to offer to upgrade they system for me from the user that actually could perform the upgrade. That’s annoying.

Use the Recovery Console! 😑

So I shut down the MacBook and restarted it by pressing and holding the power button. Holding the power button through the startup process rewards you with a notification that the MacBook with boot “with options”. It soon shows a list of options, including “Reinstall MacOS Monterey”. This was not very reassuring because I had not installed MacOS Monterey yet, but I figured that this was just in keeping with the relatively shoddy upgrade experience thus far. At the very worst, I would have to restore from my Time Machine backup after the installer had wiped everything out. Annoying, but not irrecoverable.

I started the upgrade and it soon told me that it would be done in 2h20m. That’s about 3x longer than the iMac, which is 5+ years older. What is going on? After about 15 minutes, the progress dropped precipitously to 53m. At that point, I left it and went for a long walk (several hours).

Log in and … phew! 😮‍💨

The upgrade was finished by the time I got back. I logged in and it went back to the white-apple-with-progress-bar-on-black-screen phase, and got stuck at about 90% for at least five minutes.

After that, it was logged back in to the desktop, with my user restored. Oddly, it asked me to log into my Apple Account, which I skipped, as usual, because it was going to be logged in anyway, once it had loaded my user and desktop and open windows.

Many months ago, I watched the first segment in Track 1, called Debunking “Bitcoin Wastes Energy” by Nic Carter (The B-Word). Below are my notes and thoughts about it. He started off by asserting,

“As a neutral, global monetary network, Bitcoin has a valid claim on some of society’s resources.”

Does it, though? You can try to make this axiomatic, but I’m going to respectfully ask for justification. As someone speaking at a Bitcoin conference, you obviously have to claim that. It’s good that he did. That means that if I don’t accept that premise, then any energy used by Bitcoin is wasted.

Is the utility it provides to “tens of millions of people” enough to justify its energy use? There aren’t even that many people using Bitcoin—really, for anything other than HODLing it—and it’s already using as much energy as Argentina. That doesn’t bode well as a scalable solution. It currently consumes an enormous amount of energy and does nothing useful. The claim is that we should move everything to this technology and … profit?

This guy’s logic is descriptive. He thinks that details of Bitcoin’s process are justifications for its existence. I think that the question is: can the service provided by Bitcoin be done more efficiently?

The proof-of-work is required to prevents central banks from taking over the network, defeating decentralization. But the mining is already heavily centralized (and will continue to become more so).

He claims that the major part of the energy usage is not in maintaining the blockchain, it’s in initial creation. That would at least mean that once all the Bitcoin has been created, that it use much less power to keep running. Still, I’d like to see numbers (or hear them).

Why does he keep comparing it gold? How does that actually compare? Why are numbers only estimates?

.25% of the world’s electricity? For 10s of millions of people? That is very inefficient. It uses power. It produces heat. It must be justified.

A lot of this guy’s numbers are vague and “unknowable” and “best estimates”, which is worrying…because he doesn’t seem to be worried about drawing conclusions from shifty numbers.

He discusses how tumble driers use 1.6x as much energy as Bitcoin. I assume that’s for the whole world. Then he says that,

“Of course, a tumble drier is a discretionary thing – you can dry your clothes on the line, if you want”

And, of course, Bitcoin, is not discretionary because that value that it adds is so far above and beyond that of billions of people having dry clothes. None of this makes any sense if you’re not already wearing the rose-colored crypto-glasses.

But copper and zinc and gold mining produce useful materials that are used to build other things. Bitcoin just produces … Bitcoin. It’s self-fulfilling.

“The perceived merit of an application’s energy consumption is a function of one’s subjective view of that application’s utility.”

He has, so far, said nothing about Bitcoin’s actual utility, except for some hand-waving and vague mentions of its distributed and decentralized nature and lack of overt control by nation-states and their central banks (it’s not a fiat currency). The de-facto control by central banks and large financial institutions seems unavoidable as Bitcoin achieves its goal of being more mainstream, which leads to it being absorbed by everyday capitalists.

Much of his reasoning is circular, begging the question.

Even if Bitcoin’s energy were to come from green sources, that still means that we’re producing more energy just to support it. The utility of it has not been proven—this guy hasn’t even mentioned it once. He’s once noted that you’re going to be skeptical if you think Bitcoin is useless, but not said why Bitcoin is useful.

So Bitcoin’s location-independence allows it to use otherwise wasted energy? What happens if the value of Bitcoin collapses? Then it is no longer profitable to do this — and it will stop.

Why would Bitcoin miners voluntarily spin down when demand gets too high? Even though it would be feasible for them to do so, there is no realistic scenario where they would do so.

He describes Bitcoin miners as socially aware altruists who will only use energy when it is not otherwise societally detrimental. But that’s hogwash. No-one does that. The only way anything happens on that level is through the cudgel of regulation and enforcement or through pricing, which is rarely dynamic enough to stop Bitcoin mining. If the price of Bitcoin is high enough, then many, many other more societally essential services will become unaffordable before Bitcoin mining does.

This is also ignoring that over 90% of Bitcoin production was outside of the U.S. anyway. It was mostly in China. 60% of the hashrate has left China. He argues that coal-based mining is therefore reduced. But most of the mining has since moved to Russia and India and Kazakhstan, where the energy sources are anything but clean.

According to a recent article, China’s coal-based energy is actually much cleaner than that in other locations. His argument that the U.S. grid is “much more sustainable” than China’s is not really borne out by data (he doesn’t provide anything to back up that claim). Buying offsets doesn’t help the planet; it’s a kludge available to the wealthy to assuage their consciences.

I would imagine that the recent price surge in Bitcoin is also due to media attention and subsequent popularity combined with the relative scarcity of new Bitcoin. I wrote that last bit in August of 2021. The recent drop in Bitcoin’s value will probably also make it impossible for many to be able to afford to keep mining it.

But the demise of Bitcoin has been predicted many times. There seems to be no reason for it to exist, but many people are nonetheless convinced that it should—even if they can’t articulate why.

I recently stumbled across a recommendation on Reddit from /u/schoschi1337 for reading articles that are behind paywalls (without paying for the subscription, of course). I recommend paying for the content that you read online. I pay for a lot of stuff I read online.

Paying for content honestly the only way to combat low-quality, high-engagement tripe that is otherwise the lowest common denominator of 95% of the Internet, where everyone’s hustling for the increasingly fragmented and increasingly worthless attention of hyperactive and impatient consumers trained by decades of clickbait to…wait, where was I? Oh, yeah. Pay for your content if you like it.

If you’re still trying to find out whether you like it before you pay for it—a perfectly legitimate thing to do when everything is on a subscription model these days—then sometimes you’ll have to figure out how to get into a walled garden, like a ninja 🥷.

To that end, there’s this thing called 12ft (i.e. “Show me a 10ft paywall, I’ll show you a 12ft ladder”) that lets you get around paywalls. I used it the other day and it worked perfectly. Just go to the web site, paste in your URL, click “Remove paywall”, and tada! You can also just paste your URL after https://12ft.io/ and it will work, as well.

As to how it works its magic, it claims that,

“The idea is pretty simple, news sites want Google to index their content so it shows up in search results. So they don’t show a paywall to the Google crawler. We benefit from this because the Google crawler will cache a copy of the site every time it crawls it.

“All we do is show you that cached, unpaywalled version of the page.”

There’s a ShortCut for iOS (although it’s untrusted, so “buyer” beware). I haven’t tried either the ShortCut. I thought I saw a browser extension the other day, but must have been dreaming because it’s not there anymore. Maybe the author retracted it.

So I bought an M1 MacBook Pro (teacher discount FTW!) at the beginning of October and I’m very happy with it. It’s last year’s model, not the fancy new stuff, but it’s still worlds better than any laptop I’ve ever used. Here are the things I like about it so far.

• The battery 🔋lifetime is extraordinary
• It’s very quiet 🤫
• The keyboard ⌨️ is very nice
• The touchpad is, as always, fantastic
• The Touch-Bar is quite useful
• The sound quality on the speakers 🔊 is very, very good (especially for its size)
• The screen 📺 is bright and crisp

In short, I only have a couple of minor quibbles, barely even worth mentioning. Whereas I like the Touch-Bar, I wish you could force it into function-key mode without having to press fn. That would be more useful for when I’m in IDEs. For regular use—writing, browsing, viewing, messaging, listening, etc.—the lack of function keys isn’t an issue.

The power plug takes some getting used to: I’d grown accustomed to the 2012 MacBook Pro I’d been using, which still had MagSafe technology. With that one, you can just plug the laptop in without pressing on it. With the USB-C/Thunderbolt adaptor, you have to brace the laptop on the opposite side to get the connector to plug in completely.

I’ve read that both of those issues have been addressed in the newer MacBook Pros, with the M1X and M1Pro processors. MagSafe and hardware function keys are back. I would kind of miss the Touch-Bar, though. It’s very convenient for emojis and spelling suggestions, I must admit.

On the very positive side is energy efficiency. I’ve never heard the fan, even during installations of XCode or longer compilation, which always kicked in the fan on other laptops I’ve used (even on the older Intel-based iMac I have).

Here’s what my first five full days of usage looked like, on a single charge.

Battery Usage for first six days

First charge lasted 120 hours

I let it drain right down just to see how it behaves. It’s really very, very nice not to have to always worry about the power adaptor.

The New Hotness

Since then, the new MacBook Pros have arrived and are even faster. But you’ll notice that I didn’t mention the speed of the notebook as my favorite part. It’s endurance and silence are much more important to me—because I know that it’s going to be more than fast enough for anything I need to do.

While the new machines are faster—they have more CPU and GPU cores—I noticed that they also don’t last as long on battery. I would imagine that this is due to the larger screen size. The M1 provides more than enough power for a programmer. I think the new laptops would be very attractive for gamers—if anyone were still making marquee games for MacOS. Apple always talks about how fast things render, so they’re probably targeting content-creators (artwork, videos, etc.) with their newer, faster machines. A lot of programmers are going to think that they need the speed as well, and will end up paying a high premium for it, above what they would pay for a standard M1 with a slightly smaller screen.

I stumbled across the article deMoore’s Law by Freddie DeBoer (SubStack). I follow DeBoer for other reasons, not because he is a professional tech reviewer. But, his point is that most reviews are written by people obsessed with CPU performance, which doesn’t matter to most users. As the author writes,

“I find myself skeptical that the average Macbook user is ever going to use even a significant fraction of that power. It reminds me very much of the horsepower war in cars.”

and

“What I’m not sure about is whether this matters very much: I think few people are CPU-limited in most contexts these days, and power efficiency is less important for laptops than for most other types of tech.”

These are very silly things to write. Of course power-efficiency is important. It’s a device with a battery. It is wonderful that it lasts for days without a recharge. We should all be happy when a device doesn’t need to be plugged in every night. Normally, having a faster processor that you don’t need would be a problem, but the M1 delivers top-end performance while also being very power-efficient. It does this by throttling back its power (and, therefore, power-consumption) when it’s not needed. When it is needed, it’s seamlessly there. It provides a non-stalling experience and doesn’t get in the way of what it’s for: doing what the user wants when the user wants it.

The reason that reviewers run tests with high-end editing tasks is that nothing else makes a blip on the radar for the CPU. They would all be equally fast. It would be like playing the original Doom on any modern CPU. It just runs at 300FPS or more. Or testing how quickly a car goes from 0–5kph. The answer wouldn’t tell you anything. A machine that delivers amazing high-end power will most likely use a lot less power to deliver much lower-end power, is the point they’re trying to make, I think.

That’s why Apple has their two markers of battery performance: how long can you continuously play full-screen video? And, how long can you continuously browse the web. That’s what’s relevant for most users and, in those two categories, the M1 delivers 2x to 3 longer than most other laptops you can buy.

The other day, we wanted to hike L’Argentine, a challenging day-hike in the canton of Vaud in Switzerland. It starts and ends in a little “town” called Solalex. I put the “town” in quotes because it’s really just a couple of restaurants, a tiny museum, a couple of tiny shops and a few farmhouses. It’s lovely, but it wasn’t, in retrospect, surprising that the navigation system in the car didn’t know where it was. [1]

We knew a town on the way (Villars sur Ollon), but otherwise weren’t so sure of the geography to map our own way from there. So we got out a smartphone and asked Google Maps to find a route. Why Google Maps and not Apple Maps? Because my navigator has so many apps on her phone that she couldn’t find Apple Maps right away, so she used Google Maps, which she’d found first.

Google Maps showed the following routes, with the shortest route selected by default. Not knowing that much else about the roads in that region, we took the suggested route.

Rte de la Peuffeyre on Google Maps

We drove along that route until we got to this fork in the road, with Villars clearly marked to the left (that’s the town I was nearly certain was on the way to Solalex and through which I’d ridden my bike a few years back) and Google insisting that the route lay to the right.

We should have stayed right toward Villars

From another angle, you can see that Les Plans is to the right and is also marked with a road sign that doesn’t indicate in any way that it would be difficult to get there with a car.

The right way and the wrong way

The problem is that Google wasn’t going to stay on that wide(r) road and would instead turn us onto the Route de la Peuffeyre, whose entrance looks like this.

Rte de la Peuffeyre − 24% climb in Wallis

Perhaps an abundance of optimism accompanied by an unfamiliarity with navigation systems—we’d recently second-guessed ours a few times and been proven wrong—led us to think “it won’t be so bad … I’m sure it’s just the initial rise that’s 24%.” At any rate, we began climbing and the incline did, indeed, abate, but the road was still just about as wide as our car (a 2014 Golf, not a large car by any definition).

The following screenshots are from Google Maps, proving that Google drove their little mapping vehicle over this route and came to the conclusion that it was an accessible road for vehicles.

It's pretty, but Rte de la Peuffeyre is not for motorized vehicles

Rte de la Peuffeyre is even hairier in the forest

Even once you get out of the forest, the road is persistently only as wide as a car and there is almost nowhere to pull off should you encounter oncoming traffic.

Rte de la Peuffeyre is barely as wide as a Golf

After realizing that the car (barely) fit on the road—there were a few places where the left wheel was perched quite precipitously on the edge of a road that had suffered throughout the massive storms we’d had that year—oncoming traffic eventually became our biggest concern. If this road was barely drivable moving forward and uphill, what would it be like reversing downhill along it, looking for one of the two or three (tiny) pullouts on the 1.5km stretch?

It is unconscionable that Google Maps suggests this route for cars, especially considering that Google employees have actually been on it themselves. It’s obvious that no navigation system should recommend a car drive there.

And what about Apple Maps?

Apple Maps doesn't think Route de la Peuffeyre is navigable

Apple Maps does the right thing. It doesn’t even acknowledge that there’s a road there at all.

You would think that the latest version of MacOS would work seamlessly with the latest version of iOS during an upgrade.

I had my iPhone plugged in to my Mac to charge. I noticed on the phone that it wanted to upgrade the OS, but figured I’d do the upgrade through the Mac so I could use the keyboard and mouse to navigate.

Suspiciously, MacOS did not seem to be aware that there was an upgrade pending for the phone. I refreshed the status and then it noticed that there was an upgrade and offered to apply it.

I told it to go ahead and noticed that there’s now a progress bar at the bottom of the Finder window to show download progress, which is a nice improvement in Big Sur that brings MacOS up to the HIG [1] of ca. 2000 or so.

Unfortunately, the operation now locks the entire Finder for the duration of the 4GB download, which is much more like the HIG of OS 9 than OSX or MacOS. It’s 2021 people.

Finder is locked during iPhone update

I let the update proceed and complete, but MacOS wasn’t 100% on board. The phone was fine. It had stopped complaining about an upgrade, but even after unplugging and re-plugging in the phone, MacOS still saw the phone as having the older version of the OS.

The After unplugging … still no clue it's been upgraded

Note that I’d initiated the upgrade from MacOS and it was still blissfully unaware that the phone was up to date. It even popped up a notification to let me know that an upgrade was available—the same upgrade that we had just applied together.

No idea he's just updated my phone

After unplugging and re-plugging the phone, the iPhone page in the Finder still couldn’t remember what it had just done minutes before. Instead of just showing the old version, now it was also offering to upgrade to the OS to which it just upgraded me.

Utterly unaware of what's going on

A few unpluggings and manual refreshes and MacOS finally got the memo and settled down.

As with all complaints about Apple on this blog, it’s just a mystery how Apple manages to bungle such a standard operation that is (A) executed millions of times by its users and (B) involves only their own hardware and software in the latest versions. This is just sloppy testing.

I’m a former Google Play Music user. I’ve recently moved to YouTube Music. [1], which is not very friendly for people with large playlists:

• It doesn’t pre-load the entire list; you have to scroll to see more entries
• You can’t search within a playlist
• You can’t multi-select items in a playlist, making management nearly impossible. You have to copy or move each song individually.

All of these things make managing a library a pretty big hassle. It’s like they’ve never heard of my use case before.

Anyway, I’m testing Apple Music, which addresses all of the issues above and integrates pretty well with everything else I use.

Still, there are wrinkles when using Apple Music internationally. If I click an Apple Music link in the browser, Apple detects my location as Switzerland by default, so it redirects me to the Swiss Apple Music store.

I logged in, but was rejected because I was at a store that I hadn’t signed up for. Instead of redirecting me to the US store, it showed me the following error message.

Nice translation Apple

The title tells me to “Log in with [my] Apple-ID”. The error message tells me that “Logging in to Apple Music is possible. Try again.”

I’m almost certain that there’s a nicht missing there. I can’t believe that they’d write such an encouraging message, like, “Login is possible! You can do it! Try again, buddy!”

As with other complaints about Apple, it’s hard to believe that such a large, rich company with a handful of tightly vertically integrated products used by hundreds of millions of users still has such glaring and amateurish errors.

I finally upgraded to MacOS Big Sur [1] after waiting over nine months. I’d grown quite comfortable with Catalina. My only complaint was an occasionally flaky wireless connection, which seems to be better in Big Sur.

As for Big Sur, there are a lot of smaller improvements. The control panel is nice. The Bluetooth menu is better. The notification center is better, overall.

I thought I’d hate the more rounded look, but I really don’t. I like it. I find the new style more immediately legible and usable. I mocked it as “Windows XP” based only on screenshots, but, in practice, it’s more effective than previous versions.

As noted above, my wireless hasn’t cut out mysteriously for a while—something that used to happen once a day or so, just for a second or two, but the hiccup threw video calls and other services into a tizzy.

However, there are some worrisome bits.

General Performance

The menus in both the Notes and Podcasts application are slow, taking over a second to drop down initially. It’s very strange. I just had seven windows open in the Finder and it, along with the Window Manager, was using about 70% of one CPU. The AppleSpell process sometimes pops up with 35%—even when I’m not typing anything.

Notes

The CPU usage for this relatively simple cloud-synced note-taking app is terrible. Typing in the Notes application is laggy AF. The lag in notes has nothing to do with the size of the note. It just sucks CPU while you’re typing. It’s like no-one has tested it or is even using this app to take notes. If you try to write a single sentence, it sometimes takes half a second for letters to appear. This is ridiculous. Over a month and an upgrade from 11.3.1 to 11.4 later and it’s no better.

Podcasts

The Podcasts app now also crashes every once in a while. Maybe once per day, just playing a podcast.

The Podcasts application has a default view of only 1 week. The available options are 1 Week, 2 Weeks, and 1 Month. Why isn’t there an option for All? Am I not capable of deciding when my list is getting too long? Was Apple worried that its UI wasn’t capable of handling an unrestricted list? On top of this limitation, the app resets to 1 week on each restart. Because of all of the crashing, this happens all the time.

If you re-open the Podcasts app from the crash dialog, it opens accompanied by a Terminal window that shows debug output. Running in this mode seems to fix the crashing, but the play/pause is laggy. This is not acceptable for a 2-trillion-dollar company with dozens of thousands of the world’s best software engineers.

Photos

In Apple Photos, they made a lot of improvements. But they, once again, made it more difficult to edit photo titles.

1. Use ⌥ + ⏎ to show the editor for the photo title.
2. Once you start typing, it sometimes just erases most of what you’ve typed with the filename of the photo—while you’re typing
3. Once you hit ⏎ to save changes, the focus is now nowhere instead of on the photo you’ve just edited.
4. Hitting the right arrow to go to the next photo very helpfully jumps to the first photo in the entire list. What is that?

In the 11.4 version, it looks like the focus problem might be fixed, but the random overwrite of what you’ve typed still occurs.

Notifications

The notification badges are now sticky and have little close buttons on them. These close buttons, however, disappear when you hover the first time, so you have to move the cursor off of the notification and hover on it again in order to make the button show up for realz and then you can click it.

Now Playing

The OS now keeps track of “playable” apps so that when you press play/pause on the keyboard, it activates the most recent app. Except that it doesn’t. It relatively accurately keeps track of playable tabs in a browser as well as the Music and Podcasts app.

It’s a welcome change to have a list of those queues in a central location and easily accessible. It’s just that sometimes the key combination just applies to a random application rather than to the last application you’ve interacted with—or even with the app that is currently playing something. Instead of pausing the playing podcast, it will just start playing a YouTube tab from the browser. This doesn’t seem like it should be that difficult.

Conclusion

This is all pretty basic stuff, using the basic productivity apps that Apple touts for their platform. In all cases, these are regressions from things that were working just fine in Catalina. Overall, it’s a solid release, but there are enough problems like those listed above to make you wonder who’s testing this OS? It’s been out for nearly a year and has gone through 4 minor revisions—and it still has quite a few rough edges.

Mirriad is a real company with a realtime tool that allows editors to place content into video streams. This tool can be used for anything (e.g. covering up a mustache that an actor has for another movie, as they did for Henry Cavill in a Superman movie), but their demonstration video shows how they’re selling this tool: as a way of injecting advertising and product placement in existing content.

The short video Mirriad end to end tech process (Vimeo) demonstrated their software in 2019. There is every reason to believe that they’re well beyond these capabilities now.

My initial reaction was revulsion at what they’re doing to movies. My secondary reaction is written below. I don’t want them messing with reality, but I have to admit that (A) I don’t really know what reality is and (B) they’ve been doing it for many decades already, on all levels. I mean, I know what I think reality is. I know what I think is real and fact-based, but I can’t really be sure. We’ve seen time and again how a consensus can be achieved on complete fabrications.

Once you accept that we can edit videos realistically like this, then you can no longer believe anything that you see. Can you? How do you know that the video you’re seeing is the original? How do you know that it truly reflects what happened, in the case of news videos, or what was originally intended, in the case of films?

In the case of news and information on which life-affecting decisions will be made, it’s clear that we need to get a handle on this. We’ve already seen how terribly the world acts when it’s fed fictitious information. We go to war. We hate the other. We hate ourselves. We stop taking medicine. We trust nothing. This is powerful stuff. But it’s not really new.

Putting an extra bottle of Corona in front of me is almost the least-harmful propaganda I can think of. I suppose it could subliminally encourage me to eat healthier and take more walks and smile more.

But that’s not what media does. We already have TVs positively filled with police procedurals that glorify police. There are horrifically violent movies that glorify the military every step of the way—usually the U.S. military. Content has always made minority groups look bad…now that’s changing, but we only rarely see content that sheds this legacy instead of just turning it on its head. That kind of manipulation doesn’t need any kind of technology, though, and it was powerful enough. It can always get worse, though, so I suppose if you mix it with technology, it will be even more powerful. We’ve already seen that as well.

This line was crossed in the last several decades as we accepted more and more CGI in our content. We’re used to thinking of CGI as “special effects”, like spaceships or monsters. But CGI is at its most powerful when it enhances a reality or shows a convincing reality. No-one’s going to believe that there are 60m-tall monsters walking around or that there are fleets of spaceships blowing each other up. You can make a movie about that without people thinking it’s real.

But what if you make a movie about groups of people meeting, say, under Nazi flags? For example, The Man in the High Castle is a show that is positively filled with CGI, but you notice almost none of it.

What if you suggest a movie or show like that is a documentary? What if someone does it for you? What if someone takes the content and edits your product into it? What if there’s only a small disclaimer? What if there’s no disclaimer? What if people don’t see the disclaimer and share it all over the place.

And how do we know that this isn’t already happening? Doesn’t it sound exactly like what is already happening?

We believe what we see and, even if we think “this isn’t real” while we’re watching it, our brains will helpfully stitch what we saw into memories and start to affect what we believe. Many of us will then “remember” that we saw a documentary or news clip about Nazi meetings regularly occurring in the modern day. When the news tells us that there are fascists all around us, we “remember” those images and confabulate a worldview that feels consistent, but is based on deliberate—or even just incidental and inadvertent—manipulation.

Most “news” organizations crossed the Rubicon long ago. They may have done so reluctantly, but the exigencies of competing in a purely capitalist environment drove them inexorably to it. Many didn’t seem to resist at all. They seemed almost relieved that they could stop pretending to care about actually informing people. [2]

The media are now full-blown fabulists, concerned with the “angle” instead of with presenting “facts” and “information”. Think of how they use re-enactments to show how things happened during a news show about a crime, for example. That’s all that people will remember. They know this. People will remember seeing the accused actually commit the crime. They will swear on a stack of bibles that they say it. Most people will no longer be dissuaded that they didn’t see it, unless someone can prove to them that they couldn’t have been there. And maybe not even then.

Since the time when Plato pointed out his cave shadows or Quine his qualia, we’ve had to deal with the notion of having sensors and interpretive equipment of dubious and highly varying quality. That’s bad enough, of course, but there’s nothing that can be done about it. We have to live with what we have in the way of eyes and ears (primarily; I know touch and taste and smell enter into it, as well, often very powerfully) and try to control our filter by (maybe) writing things down and discussing them and vetting them.

There are some who’ve doubted images even back when we didn’t have the power to fake them. For example, the Moon-hoax conspiracy theorists don’t believe we’ve ever been to the Moon. There are eloquent debunkings (e.g. Moon Landings Faked? Filmmaker Says Not! by S G Collins in 2013 (YouTube)) of this conspiracy theory that appeal to the fact that, in the 60s, we didn’t have the computing power or the film stock to fake all of the video we saw. We just didn’t have the capability to lie to ourselves at that level in film. It was actually more feasible to build a ship to go to and land on the Moon than it was to fake the videos of it.

We can no longer think that. We can fake nearly anything now. Deep fakes (from 2021) are just the beginning of how we can be manipulated. There are also deep fakes of fake people to worry about, as well.

So, if Mirriad is using their technology only to place products in old movies, that’s the least bad thing they could do with it. At least we can detect that, somehow, if we compare to a verified original. It’s not harmless, but it’s less harmful than fake videos of fake fascist leaders of organizations that don’t even exist, all designed to drum up fear. When Trump called everything “fake news”, he was—as he occasionally was—right for the wrong reasons.

Organizations ranging from obscure and niche web sites to giants like Newsweek collaborate to invent a Chinese enemy (from 2020). Will we even notice when they start using this technology to “improve” their version of reality? How do we know they aren’t doing so already? The set of tools they’re willing to use has grown over the years. With each year, it becomes “normal” and “accepted” to “fudge” things so that they’re more convincing, so that they’re more likely to convey the “truth” they’re supposed to convey.

Articles are edited after publication, with no correction noted. Entire videos are partially redacted in order to rob context that would lead to unwanted interpretations. Certain voices are amplified while others are suppressed. And always lies, lies, lies, many told with no other goal than short-term financial gain. There is no consideration for the damage done to society.

But this happened—and continues to happen—without special technology. It’s pretty much as old as time. William Randolph Hearst was whipping up a war way back in 1898 with nothing more than a single popular newspaper organization. [3]

The media is vastly more powerful today and can disseminate information faster and farther than ever, but the concept hasn’t changed. People haven’t gotten smarter. They are still super-easy to influence—perhaps especially politicians with power over so many lives. They are no better-informed than anyone else, unfortunately. Some are. Most aren’t. Many, if not most, end up believing that which enhances their own personal wealth the most. Technology has nothing to do with that.

I recently followed a link to a video on YouTube that was supposed to demonstrate the madness of Japanese wrestler Jun Kasai. I was unable to view the video because it was age-restricted and I had not yet verified my age for my Google account. [1]

Google helpfully offered a couple of ways of rectifying this issue.

Google's Age-verification Options

That escalated quickly.

I want to ask incredulously “do people really scan their driver’s license and send it to completely unanswerable company like Google?” or “hey scan and send your passport, why not?” … but I already know the answer.

People want to see a Japanese wrestler destroy himself with fluorescent light-bulbs or they want to see boobies and they will not hesitate to give away the information anyone could use to spoof their identity.

Or I suppose you could let Google have your credit card, instead. All of these options are good.

Apple bent its rules for Russia—and other countries will take note by Lily Hay Newman (Ars Technica)

“Questions remain about whether Russia’s end goal is to completely isolate and disconnect its Internet from the wider world or whether the government prefers a hybrid network. But from the Kremlin’s perspective, the opportunity to promote certain apps on iOS is a boon either way.

“Apple could have simply allowed Russia to pre-install whatever apps it wanted on iOS devices, but the company also could have taken a radical stand against such interference. Instead, it found a middle ground, one that other countries may well seize on to suit their own autocratic interests.”

This is written by a woman who lives in the country that already controls all of the companies that make all of the apps on every phone around the world. It’s sad that she can’t even imagine why another country would think it’s not a great idea to let that stand unchallenged, especially when the U.S. is so heavily politicizing its software right now.

Those dastardly autocratic Russians—forcing U.S. companies to allow them to ask their citizens if they’d like to install non-U.S. propaganda on their phones. Obviously, no Russian will install state software, though, right? I dunno. Probably Yandex? Or Telegram?

She did later write:

“From both an economic and national security standpoint, it’s understandable to a degree that governments would want to promote domestic software to their own citizens. But in practice, the Internet’s growing balkanization is eroding Internet freedom worldwide and undermining the entire concept of a decentralized, global web.”

“Balkanization” meaning the 100% control by U.S. firms is eroding somewhat. But not much. But enough that it can elicit pants-shitting terror and projection on an international scale, accusing the Russians of trying to do what we’ve been doing all along.

Just listen to how this is written:

“The situation with Russia’s mandatory apps is not the first time Apple has faced invasive legal requirements from an authoritarian government—nor the first time the company has conceded to these demands.”

This from a woman living in a country that is considering broadening the already-existing requirements to force backdoors into secure software. A country whose NSA already did that and continues to do that all the time. In Europe, they’re trying to do the same thing. Again. But these efforts are called “promoting democracy” instead of “authoritarian”. It matters who’s trying to do it, right? And that’s not even what Russia is doing: they just want Apple to ask their users if they would like to install some standard Russian software. They can just say no and continue on with their day. Nothing about making people worldwide do it, nothing about backdooring existing software, nothing about changing iOS.

“[…] the suggested apps aren’t pre-installed, and users can opt not to download them.”

That sounds pretty autocratic.
 

The article 2021 Prediction #4: WiFi 6 is a bust (for now) as Bufferbloat returns thanks to ISP greed by Robert X. Cringely takes up a topic that he’s written about quite a bit over the last decade: bufferbloat. This symptom of a bad connection shows up as excessive buffering and excessive latency when streaming video (or a lot of data). Basically, the connection will continually fail and restart until it stops altogether as the router gets all snarled up. The following definition is from Wikipedia, which has even more technical information.

“Bufferbloat is a cause of high latency in packet-switched networks caused by excess buffering of packets. Bufferbloat can also cause packet delay variation (also known as jitter), as well as reduce the overall network throughput. When a router or switch is configured to use excessively large buffers, even very high-speed networks can become practically unusable for many interactive applications like voice over IP (VoIP), online gaming, and even ordinary web surfing. (Emphasis added.)”

The problem has basically been solved. It was solved soon after it was discovered, but the problem since then has been distribution. ISPs are still chirpily distributing routers with “excessively large buffers”, which means that it doesn’t matter what your upper limit on bandwidth is—your router isn’t going to be able to use it. As far as I can tell, though, Cringely is mostly talking about the United States when he writes that “hardly any ISPs support” proper connections.

“All of the anti-bufferbloat internet standards mentioned here have been published and available for at least eight years, yet hardly any large ISPs support the new code in their default equipment. I can only guess that’s because it solves too many problems and lessens demand for bandwidth. Think about it, ISPs make a killing selling you excess bandwidth you don’t actually need. Electrons you aren’t using don’t exist and therefore those bandwidth upgrades are 100 percent profit. (Emphasis added.)”

Unfortunately, the ISP has zero incentive to fix bufferbloat and a huge incentive not to fix it. The only thing standing in their way is basic ethics, morality, and the desire to provide a good service before making a bigger profit. Cringely goes on,

“If you don’t know if you have bufferbloat, you are lucky. Check by running the DSLreports speed test, NOT the speed test from your ISP.”

I can confirm that DSLreports speed test is quite good and shows you much more data than my previous default Speedtest.Net.

My connection is healthy

Cringely recommends the following:

“Users are much better-off staying for now with 802.11n or ac at the mid tier of Internet speeds with good bufferbloat fighting tech if they want to sit at home sharing the link with their kids.”

The problem has come to the fore again because everyone is running video calls from home all day and their already-trying experience is made worse by a stupid configuration error in the firmware of their routers.

If you run the test above and you have a bad bufferbloat score (the average in the U.S. is C+), then see if you can get a router that Cringely recommends (he lists several) or ask your ISP if they can update the firmware. Do not upgrade your connection; it almost certainly won’t help and would be throwing good money after bad.

Cheyenne 9’s Coverage of the Official White House Address by Sassy Justice (YouTube)

Just in case you think you can spot a deep-fake video from a mile away, the gentlemen that brought you South Park made a new Sassy Justice video. They do us the favor of making it pretty clear that this never happened; others will not be so generous. On the surface, it’s just a joke video, but the implications are grave. If Sassy Justice can make this video, what’s stopping a much-better funded news organization from making a “Trump/Biden calls staffer the c-word, caught on tape” video?

The SMBC comic about villainy shows a master criminal who “can generate a perfect deep-fake of any public figure doing anything”, to which his interlocutor responds “Why waste all that computing power? Just take an out-of-context still of the politician you don’t like and writer whatever claim you want on it.”

“Using neural networks to fool the human mind is like using a bazooka to fight a hamster.”

Install the COVID app, they said.

It’s for the good of the nation, they said.

It won’t drain your battery, they said.

Bollocks.

COVID takes 60% just sitting there

The yellow color indicates that I had the phone in low-power mode.

The phone was on my desk all day long. I used it twice for 2FA. I looked at a few messages.

I’d last charged it sometime yesterday evening.

I wonder how much battery it needs when it’s actually doing something?

iOS has long had a setting that lets you decide whether “Background Refresh” is enabled. I have always had it disabled because I don’t need any of my apps doing things when I’m not using them. I’m more interested in my phone being there for me to look up something useful than I am in looking for a charger or carrying a power brick wherever I go.

This does not mean that apps are not doing things in the background, though. The article Background App Refresh Explained by David Johnson on Dec 6, 2019 (Techzillo) explains that the “background” state means that,

“[t]he app is in the background and executing code. Most apps enter this state briefly on their way to being suspended. However, an app that requests extra execution time may remain in this state for a period of time.”

When “Background App Refresh” is enabled, apps are given more leeway to do things when in a mode other than Active. The article Updating Your App with Background App Refresh (Apple) provides more information on how applications hook into this behavior and set fetch intervals and perform work.

That all sounds well and good, but I’m still seeing some apps running absolutely amok on iOS 13, seemingly ignoring all of the background settings and churning through heroic amounts of energy and consuming hours of CPU, all without my using the phone at all.

It’s not active. There is no app that is active.

Obviously, there is an app in the foreground, but the phone is off and lying on a desk. I have left no operation running other than leaving communication clients open. They are notified of messages through Apple’s centralized push-mechanism, which is very efficient.

The other day, my battery was flat and I could not imagine what had done it. It was Shazam.

Shazam used 15 hours in the background

Just to be clear, Background App Refresh was disabled.

Background Refresh is Off

Also, Shazam reflects this in its own settings.

Shazam isn't allowed to refresh in the background

Shazam doesn’t care, though. Shazam gonna do its own thing.

Apple acquired Shazam almost two years ago. All the app is supposed to do is listen to a song and tell me what it is. What it instead does is it manages a completely separate song list from Apple Music as well as starting to play completely unasked-for videos.

I looked up one song and then made the mistake of not killing the Shazam app with extreme prejudice.

I imagine it burbled its way through about 15 hours of playing videos and looking shit up and just launching threads that waste time for no conceivable reason. At any rate, without “Background App Refresh” enabled, it consumed 15 hours of time in the background. The setting is, for all practical purposes, meaningless.

I’d be tempted that this is just another way that Apple gives its own apps special permissions, but I’ve seen this same behavior with Garmin Connect and other apps. I wrote about it in iOS Battery Drain in 2017, where the Clock used up over 40% of my battery and the Home Screen also managed to use 100% of the battery at one point, nearly draining the phone. At least you can now tell which application iOS thinks used the battery; in the old days, things were more mysterious.

At this point, if you want to reliably keep your phone alive for a longer period—if you’re silly enough to want it to be there and usable when you need it, rather than having it fritter away its entire battery doing shit you never asked for—then Airplane Mode is really your only reliable friend.

Killing apps manually is a pain and hard to remember to do, but it would probably help. It’s like being back on Windows 3.1 with cooperative multi-tasking.

Disabling “Background App Refresh” might be preventing even more horrific battery drainage, but it doesn’t stop nearly enough of it. Apps are still seemingly free to go wild often enough to be an absolute nuisance.

In the aftermath of 9-11, biometrics and, in particular, bio-imaging software companies enjoyed a huge surge in valuation. Most of these products were shoddy and didn’t deliver on even a reasonable fraction of their promise.

That didn’t stop legislators from passing laws requiring their use—and probably getting giant kickbacks from companies newly flush with cash derived from their increased valuations caused, at least in part, by these same laws. Life is quite easy for some companies—especially in the world of security theater.

The article Thermal Imaging as Security Theater by Bruce Schneier lets us know that, after COVID-19, “thermal imaging software” appears to have emerged as the new darling of the security-theater world.

However, the technology has a certain niche in which it makes sense: close-up, individual readings. That is, of course, not how they’re going to be used.

Schneier points out that:

• They are not intended for distance from the people being inspected.
• They are “an imprecise method for scanning crowds” now put into a context where precision is critical.

Using thermal-imaging for use cases for which it is completely inappropriate will lead to “false positives, leaving people stigmatized, harassed, [and] unfairly quarantined” and will also create false negatives, as people exhibiting mild symptoms—or not presenting a notable fever—are missed entirely.

That a security measure is completely inappropriate, ineffective, and possibly actively counterproductive won’t prevent it from being enshrined in law the world over, though.

Contact tracing, or just “tracing”, is a way of determining who’s been infected with a contagious disease in a community. Compared to self-isolation en-masse, it’s a finer instrument: instead of everyone staying away from each other, properly trained workers trace the path of the disease, using this information to isolate the ill from the still-healthy.

How it works

When someone tests positive for a contagious disease, tracers interview them to find out where they’ve been and who they’ve “contacted” since they most likely became contagious. This period is estimated by what we know of the disease in question and the stage of infection they’re in, based on the symptoms.

For example, assume that we think that an infected person is asymptomatic for up to a week before they exhibit symptoms. Assume someone exhibited symptoms for two days and self-isolated before going to the hospital or calling a doctor. The contact-tracing team needs to find out who that person contacted for the week prior to showing symptoms. Two days can presumably be accounted for as self-isolated—except for cohabitants, who must be traced next.

They ask questions, get names and then find those people, run tests on them and fan out their interviews from there, until they can confirm that they’ve hit a dead-end on finding people who test positive.

And here’s the key: once you’ve found people who are positive, you have to isolate them to prevent them from spreading the disease further. Tracing is nearly useless if you don’t isolate based on your results. If you don’t, you just end up starting all over with tracing—back to square one.

Tracing is established epidemiological groundwork and it is very manpower-intensive. There is nothing magical about it. It’s not foolproof, but if done assiduously, it’s pretty damned good. It’s how China and Korea stamped out their first wave of Covid-19. Also because they paired tracing with isolation. Just tracing means that you’re always behind the further spread and are observing the trail rather than getting ahead of it and putting in a firebreak.

The article A Democratic Virus by Alex Gourevitch (Jacobin) gives a summary of the measures in place in South Korea and Taiwan. [1]

“South Korea, the much-hailed model, locked down two cities, currently takes everyone’s temperature in public spaces, monitors every person’s movement through cell phone and television data, and uses government and public surveillance to keep tabs on any individual suspected of carrying coronavirus so it can enforce self-isolation. Taiwan is the same. Locals report getting a knock on the door from the police a half hour after their cell phones died because their movements could no longer be tracked.”

The siren call of technology

Contact-tracing ProposalWe hate the sound of “manpower-heavy” [2] because that sounds like a lot of organization and UGH… a lot of work, so we want technology to solve this for us, just like it’s so wonderfully solved everything else.

China and Korea did benefit from technology in that they have a pretty widespread network of social-app data to which the government has access. In addition to the interview—and they did the interview parts too—they were able to use location and contact data to establish connections that might have otherwise gone undiscovered.

But it works without apps, as well, just to be clear. The cartoon to the right (obtained from Contact Tracing) explains how such a hypothetical app might work to fulfill the role historically played by people.

Will it be voluntary?

One argument for this type of technology is that most people have already willingly given up their data for the dubious benefit of partaking in various social-networking endeavors, so they shouldn’t be too troubled by an app that will literally save lives.

This isn’t a terrible point to make, but we should think about a few facets of the argument that are only implied. Social-networking apps have, among others, the following characteristics:

• Participation is voluntary.
• It is also, in theory, revocable.
• It is possible to participate anonymously (or at least semi-anonymously).
• One can have multiple identities

Will installing, granting broad permissions to, and keeping a life-saving app on at all times also be voluntary?

Can app-based contact-tracing even work if participation is voluntary? No, it cannot. It will not work at all without proper, boots-on-the-ground contact-tracing and without a high level of participation. Otherwise, you can just save yourself the trouble of developing it.

Which takes us to the next section.

What about data-privacy laws?

If you read the cartoon above, you have an idea of how the app might work. It sounds doable but, putting aside the basic feasibility (for now), how would such an app integrate into a society with data-protection laws?

• Will the data be untraceable, as the cartoon hopes?
• Which government agencies get access to the tracking data?
• When is it deleted? Two weeks? A month? After a year? Never?
• Are we issued a device or are we expected to mix the government surveillance app on our private devices?

Those are just the policy questions. Let’s assume, for now, that whoever issues the app can provide enough guarantees about anonymity and data-privacy to satisfy local laws.

What does the workflow look like?

Given that the app can detect and notify contacts, how would the whole “reporting and responding” part work?

• Do you trust a potential contact from an app enough to self-quarantine?
• Or do you have to report for a test?
• Do we trust the test enough to self-isolate?
• What about a negative result? Does that mean you’re in the clear?
• If you can trust neither a positive nor a negative result, then what’s the point?
• Is it only for the authorities to ping you to report for testing because they think you’ve contacted someone?
• Do you get a message if you rode on the same train car as someone who tested positive?
• Or does the app only work for when you’ve tested positive, to allow the authorities to track your contacts afterwards?
• Do the police show up to enforce isolation?

It’s already not so easy to strike a balance between public safety and civil rights during a pandemic. The important parts involve solidarity and trust in the government. It’s not clear to me how an app would help.

Leverage existing infrastructure

Why do we even need a new app to surveil ourselves? Aren’t we already being watched six ways from Sunday as it is?

There are two surveillance networks. One is the global U.S. surveillance system exposed by Edward Snowden. Snowden’s revelations led to nearly no change in behavior. The program was certainly not dismantled. It still exists and it has likely been extended since then.

We will almost certainly not make use of this network to track people because of … drum roll … national-security reasons. The U.S. will go to its grave with those words on its lips.

When people claim that the citizenry has already opted for pervasive surveillance, they are referring to the private networks set up by Google and Facebook and Apple, who are basically able to track most people’s every move. They did it to sell ads, but now it’s time to use it for good. I’m a bit too cynical to think that this could possibly work as, for example, the article We Need A Massive Surveillance Program by Maciej Cegłowski (Idle Words) does.

He argues (as above) that, for better or worse, we already have a surveillance system in place and it would be “shameful” not to use it for this glorious cause.

“Of course, the worst people are in power right now, and the chances of them putting such a program through in any acceptable form are low. But it’s 2020. Weirder things have happened. The alternative is to keep this surveillance infrastructure in place to sell soap and political ads, but refuse to bring it to bear in a situation where it can save millions of lives. That would be a shameful, disgraceful legacy indeed.

“I continue to believe that living in a surveillance society is incompatible in the long term with liberty. But a prerequisite of liberty is physical safety. If temporarily conscripting surveillance capitalism as a public health measure offers us a way out of this crisis, then we should take it, and make full use of it. At the same time, we should reflect on why such a powerful surveillance tool was instantly at hand in this crisis, and what its continuing existence means for our long-term future as a free people.”

So his message is pretty mixed, reluctantly coming down on the side of “let’s use this for good, but also think about how bad it is and shitcan it the moment we don’t need it anymore.”

This is where the argument bogs down: measures of this kind have historically lingered pretty much forever. The Espionage Act of 1917—originally meant to keep a boot on the necks of those upstart Bolsheviks with their dreams of socializing the world—is still on the books and has been used more than ever in the 21st century.

The Patriot Act is renewed like a Readers Digest subscription—unthinkingly and, for nearly 20 years now, largely unread. Echelon is still in place. The whole NSA network is in place. Google and Facebook gather scads of data on us, despite assurances to the contrary.

This happens all the time

The article We’re not going back to normal by Gideon Lichfield (MIT Technology Review) likens the coming surveillance and tracking measures to the concessions we made in the last 20 years in order to be able to fly.

“There would be temperature scanners everywhere, and your workplace might demand you wear a monitor that tracks your temperature or other vital signs. Where nightclubs ask for proof of age, in future they might ask for proof of immunity—an identity card or some kind of digital verification via your phone, showing you’ve already recovered from or been vaccinated against the latest virus strains. We’ll adapt to and accept such measures, much as we’ve adapted to increasingly stringent airport security screenings in the wake of terrorist attacks. The intrusive surveillance will be considered a small price to pay for the basic freedom to be with other people. (Emphasis added.)”

The analogy to flying is appealing at first. The surveillance and security-theater hassles associated with flying were optional, though. What we’re talking about is a significantly changed society. It might not be avoidable in that too many people will suffer and die if we don’t restructure, but we should at least think about each step instead of just listening to the loudest, most-panicked voices. We should at least have some reassurance that the measures being promoted will have the desired effect. What’s the point of giving up a freedom for nothing in return?

The article Protecting Civil Liberties During a Public Health Crisis by Matthew Guariglia And Adam Schwartz (EFF) asks some of the right questions.

“But we must not lose sight of the great sensitivity of the personal data at issue–this data paints a clear picture of the travel, health, and personal relationships of airline passengers. EFF would like the CDC to explain what it will do to ensure this sensitive data is used only to contain communicable diseases. For example, what measures will ensure this data is purged when no longer helpful to contact tracing? Also, what safeguards will ensure this newly collected data is not used by police for ordinary crime fighting, or by ICE for immigration enforcement?”

These are good questions that should have good answers before we proceed with anything invasive.

The article Protecting Openness, Security, and Civil Liberties by Cindy Cohn (EFF) starts by conceding that we’ve already upended our lives in many ways—but should still be careful with every additional concession.

“We know that this virus requires us to take steps that would be unthinkable in normal times. Staying inside, limiting public gatherings, and cooperating with medically needed attempts to track the virus are, when approached properly, reasonable and responsible things to do. But we must be as vigilant as we are thoughtful. We must be sure that measures taken in the name of responding to COVID-19 are, in the language of international human rights law, “necessary and proportionate” to the needs of society in fighting the virus. Above all, we must make sure that these measures end and that the data collected for these purposes is not re-purposed for either governmental or commercial ends.”

Finally, the article Privacy vs. Surveillance in the Age of COVID-19 by Bruce Schneier also encourages using every weapon we have to prevent needless death and suffering, but to recall invasive tools the minute we don’t need them anymore.

“I think the effects of COVID-19 will be more drastic than the effects of the terrorist attacks of 9/11: not only with respect to surveillance, but across many aspects of our society. And while many things that would never be acceptable during normal time are reasonable things to do right now, we need to makes sure we can ratchet them back once the current pandemic is over.”

Accept necessary but temporary measures

Just because we’re in “hammer” mode to buy time right now doesn’t mean we have to continue to act in panic. We’ve bought ourselves some time to think. We should use it wisely.

We can only hope that we get answers and assurances that the efficacy of contact-tracing measures will be continually evaluated. There have to be concrete “pull the plug” clauses in any laws that go into effect. These clauses should apply both to addressing situations where measures are being abused and also to “sunsetting” measures that are no longer needed.

If the involved parties fail to comply, there must be punishment.

Reality bites

Just kidding: obviously there won’t be.

They’ll write shitty apps that steal all of your data and do terrible, stupid things with it. The more disadvantaged your various identity groups, the worse it will be. The usual suspects will come out on top with all of the money.

Just as Schneier can’t possibly believe that his pleas won’t fall on deaf ears, I’m also deeply skeptical that our current civilization can do this. Still, I was surprised to see how much of the world enacted these isolation measures, so let’s wait and see.

Even intelligent and clear-thinking grudging proponents of using surveillance (like Cegłowski above) spend ¾ of their time hedging about misuse. History shows that power will be misused and that the laws granting it will never be repealed voluntarily.

Snowden and Greenwald on Surveillance

For more on the existing surveillance systems, the following interviews and discussions with Edward Snowden are quite illuminating:

CPH:DOX online by Edward Snowden on March 23rd, 2020 (YouTube)

System Update With Glenn Greenwald − Edward Snowden, Andray Domise and Cassie King (YouTube)

I found the second interview quite good and have included a partial transcript of the parts I found most relevant.

“Edward Snowden: Presumably—and this was the presumption of the question put to me before—is the idea that this is a choice between mass surveillance or the just completely uncontrolled spread of an infectious virus that can cause serious disease. And I don’t think that’s accurate. In fact, I know that’s [not] accurate. I know a little something about how surveillance works here.

“What we are being asked is to accept an involuntary mass surveillance in a way that has never been done before at this scale, in the context of a real crisis. They just go: ‘Look, we’re going to do this, the data already exists, the phone companies … we’re going to apply it to sort of a new use case. We’re going to this surveillance infrastructure that already exists, or rather we’re going to take this communications infrastructure that was not designed for surveillance—or, rather, it was told to us that it would not be used for surveillance—and we’re going to use it for precisely that, but for a really good reason.‘

“Now, they say that this is necessary, they say that there is no alternative, they say that if you want to save lives, you’re going to have to do this. But that’s not true. The question here is between the involuntary surveillance of everyone that has been carrying a phone over the last however-many weeks or months or years that they want to look back to. Because, remember these record of your movements, at least by AT&T and Verizon in the United States, are reputed to go back to 2008. Everywhere your phone has traveled since 2008, they know.

“There are no laws regulating how long they can retain this information, in large part in the United States.

“Now, imagine an alternative: you go to the hospital, you are diagnosed with an infection and the doctor goes: ‘it would be really helpful for you to be able to voluntarily share the movements of your phone.’ So you go in with your app and you show them: ‘I was sitting next to a guy who I don’t know who they are, but you said they were infected.‘ You now get priority access to this kind of testing, you can get priority access to treatment because it is clear that you have potentially been exposed.

“And none of this requires privacy sacrifices, none of this requires any sort of involuntary or intrusive violation of rights. And the funny thing is, these capabilities are not difficult to create. This platform could have slapped together in four days by a bunch of university researchers working together, if they had had the kind of funding, mandate and support.

“Glenn Greenwald: A lot of your answers are predicated on the desirability not of government coercion but of voluntary conduct. That is not only in the individual’s enlightened self-interest, but in the interest of society, which, in turn, means that there is a flow of information that is accurate and reliable and trustworthy, that people put their faith and confidence in, as kind of reliable font of authority to form their understanding of how the pandemic functions.

“And, maybe, I’m not sure, but I suspect it’s the case that there are countries in which there is faith in some kind of centralized authority, whether it’s scientists with government or media outlets that they trust, to get this information and it can be effective. But, in other countries, and certainly in the United States and it’s true here in Brazil and it’s definitely true throughout Western Europe, there’s a collapse of trust in these institutions of authority, where people aren’t sure anymore what to believe.”

It’s not “true throughout Western Europe”. Perhaps in France or Italy, but in Germany and Switzerland, at least, and most likely also Scandinavia, there is considerable support and trust in the government that they are doing the right thing.

If we do concede liberties to a government that we don’t even trust, it should at least be for something that actually works for fighting the disease. Can an app even do what people think that it can? We’ve already seen that the app would be just an aid to other, more traditional contact-tracing measures.

The Technology is not there

When many people hear “app”, they probably just think they’ll install it, make sure Apple’s “Health” app is running, grant all permissions to everything, turn on Bluetooth, turn on location services, and then they’ll be saved from Covid-19. No wonder they can’t wait to get their hands on it.

It’s not going to be like that, at least not for a long while. There are many technical, technological, and process hurdles to jump first.

I’ll start with another security-specialist article, Me on COVID-19 Contact Tracing Apps by Bruce Schneier, which reflects my own misgivings on apps for contact tracing.

“I’m not even talking about the privacy concerns, I mean the efficacy. Does anybody think this will do something useful? … This is just something governments want to do for the hell of it. To me, it’s just techies doing techie things because they don’t know what else to do. […] It has nothing to do with privacy concerns. The idea that contact tracing can be done with an app, and not human health professionals, is just plain dumb.”

WhatApp bleeds an iPhone dry in 5.5. hoursSnowden mentioned above that an app “[…] could have slapped together in four days by a bunch of university researchers working together”, but I think he’s drastically underestimating the effort involved, how difficult it is to develop software at scale. Not only that, but the countries currently discussing apps are also vastly overestimating the capabilities of BlueTooth, NFC, software, and battery life.

Does that damned thing even work? Can it even be made to work reliably? It depends on BlueTooth, right? Is the location data precise enough to say anything about contagion? The image to the right shows my iPhone 6s having been bled dry by the venerable WhatsApp in just 5.5 hours of doing nothing. [3] I remember years of Apple’s message app screwing up the order of messages and double-sending to some devices. It still doesn’t know what to do with non-Apple devices (the SMS barely works in groups).

Analyzing Garmin Connect

Instead of picking only on WhatsApp, let’s take a look at how well-established, years-old software working with dedicated hardware works. This example is from a Garmin ForeRunner 435 with the Garmin Connect app. It deals with health data.

The following screenshots are from the app just a few days ago. The only thing correct on it is the number of steps. Everything else is laughably wrong. Will the new contact-tracing app do any better? Or will it, too, be filled with guesswork and official-looking but ultimately fantastical data?

Garmin Stats are nearly all wrongIntensity Minutes are all wrong

The watch claims that my high heart rate was 153bpm, but I know it never went that high during the whole walk. In fact, I caught the readout on my watch as I walked into my apartment—and it read 153bpm as I was just standing there. So now it takes the misreading of my heart rate—something you would think would be working reliably by now—and extrapolates all sorts of madness from it.

For example, those intensity minutes. It looks impressive, but I know it rained most of the week and I didn’t go for any runs or longer walks with the watch. Let’s look at the intensity minutes in the next graphic: It’s claiming that of the three hours that I walked today was in a high-intensity zone for double points and 360 total minutes. That’s flat-out bullshit. It was just a walk.

Also, I have no idea where he got the 240 (4 hours) of intensity minutes on Tuesday. I must have worn the watch, but it probably just locked its heart-rate reading at 130 while I was sitting at my desk. On the other hand, I wore the watch for a very intense hour-long workout on Saturday, which doesn’t show up at all.

I walked about 12km today with about 400m of incline, but my watch thinks I only burned 2,222 KCal all day, which seems like it would be a normal amount for a day on which I hadn’t walked 12km.

I don’t wear my watch when I sleep, so the app is 100% guessing on how much I sleep. It consistently guesses between 10.5 and 12 hours, which is off by a good amount. I almost never sleep longer than 8 hours. I know that’s wrong. The app is literally pulling numbers out of thin air.

Trusting false technological Gods

Let’s imagine an app developed as quickly as possible. Tight deadlines do not a reliable app make. WhatApp is an old, established app maintained by one of the best developer teams in the world [4] and even that thing occasionally just drains a phone for no reason. And WhatsApp doesn’t even use BlueTooth or Location Services. Garmin has also been making apps and devices for a long time and their stuff has gotten progressively better. But their stuff isn’t even accurate enough for a hobbyist to use for tracking health data. How is a first-time app going to be more reliable than these apps right out of the gate?

So what’s the problem? The problem is that these apps have been in development for years and their data can’t be trusted, even for fun. The problem is that we then expect a brand new app to do a better job just because we want it to. And this new contact-tracing app is serious business. When it gets its data wrong, people get sick and people die. The problem is that people imbue these devices with God-like awareness and accuracy. Once numbers exist, they must be correct, … right? Right?

We don’t know how to write reliable software [5]

Software is notoriously hard to get right. Hurrying makes it worse. These apps will most likely be developed by amateurs (e.g. Swiss college students are, by definition, amateurs) or jaded professionals, all steeped in a culture of building MVPs (minimal viable products) and using iterative release pipelines to fix things up in production, using customers as beta-testers. Are people aware that they’ll be entrusting their lives or future health to beta software?

The people who know how to build apps have no practice building reliable, bullet-proof software. Their motto is “fail early, fail often” and they iteratively get better over time. Sometimes. Often enough, software hits a local maximum that isn’t very high at all, ending up in the doldrums of “good enough”. We don’t really know to build software that is great and reliable from the get-go. [6]

The first versions will be atrocious and will most likely torpedo the whole effort. Just ask the SBB (Swiss Railway System), which took years to regain its reputation after a catastrophic nationwide app release. The contact-tracing app will be built by people even less professional than that.

It’s not that it’s impossible, just that it’s highly unlikely. Until it does get reliable, though, decisions will be made based on its shitty data-detection and its incorrect algorithms.

Here come the trolls

And, even then, even given that the app works as advertised: it’s fast, accurate, low-power, and guarantees anonymity, what about misuse? The article Apple and Google detail bold and ambitious plan to track COVID-19 at scale by Dan Goodin (Ars Technica)

“Another possible weakness: trolls might frequent certain areas and then report a false infection, leading large numbers of people to think they may have been exposed. These kinds of malicious acts could be prevented by requiring test results to be digitally signed by a testing center, but details released on Friday, didn’t address these concerns.”

Kids could get out of school by sending in a kid with a phone that alerts as positive. The whole school closes. They’re all quarantined. You could tie your positive-bleating phone to a dog and have it run through a neighborhood. Or you could send it on a drone through a mall. There’s nearly no end to the mischief that unserious people could use to abuse the system.

tl;dr

So where does that leave us?

• Contact-tracing is manpower-intensive, but works without an app.
• Building and deploying an app is not as easy as people are making it sound. It’s not a panacea.
• It’s not clear how an app would help rather than hinder.
• It’s not clear that an app can even be made to work as hoped.
• Even if it does, trolling will be rampant and threaten to ruin trust and data.
• We’re terrible at building reliable software. We’re even worse at it when we hurry.
• The first iterations will be nearly catastrophically bad—each of which could torpedo the whole project by killing people’s trust in the app.
• Once we’re surveilled, we’ll probably always be surveilled. Power doesn’t relinquish power willingly. If we accept an app, then it, and its data-collection, is here to stay.
• That data will be put to other, non-pandemic uses sooner rather than later.
• Those uses will not be to your benefit.
• Trading privacy and liberty for an app that doesn’t have much of a chance of delivering enough compensatory value is a bad deal.

In the Storage Manager [1] on MacOS, there are several categories on the left-hand side, each showing the amount of space used for that category. I noticed that, for whatever reason, I still had the GarageBand base files installed. I haven’t ever used this application, so I had no qualms about deleting these files to recover 2GB of space on my SSD.

When I clicked “Remove GarageBand Sound Library”, the system asked for administrator credentials.

I entered my credentials but the credentials prompt stayed on the screen without change.

That’s odd.

Maybe I’d entered them incorrectly and had missed the classic “head-shake no” UI gesture.

I entered them again.

Same thing. The credentials just cleared and the box stayed on screen.

That’s very strange. Unsure of what’s going on, I decided not to keep entering my administrator credentials into what seemed to be malfunctioning software. I canceled the dialog.

Same thing. The box stayed on screen, even after canceling.

I canceled again.

Same thing. Unchanged.

Curious, I dragged the window a little bit, to be able to see the underlying “Storage Manager” dialog in its entirety.

There was another credentials prompt behind it.

And another behind that.

I kept dragging them apart until I found six credentials prompts open at the same time.

Many Credentials Prompts in the Storage Manager

Now I knew what was going on: The files in the “Music Creation” category were in several folders, each of which was individually protected and need elevated permissions to remove.

Instead of collecting permissions, elevating access, deleting all of them at once, and then dropping privileged access, the operation ran through each step for each individual folder. It does this in parallel, which is why all of the dialogs show up at once, overlapping one another perfectly.

Now that I knew what was going on, I entered my credentials six more times. As each dialog closed, the number displayed next to “Music Creation” dropped by a bit, until the category disappeared entirely when I’d entered the last credentials.

There were probably about nine or ten dialogs initially, so be patient if you want to delete your “Music Creation” files.

It as strange, initially worrying, and inefficient, but at least it worked, in the end.

Why are multi-language web sites so hard to make? Even large companies like Microsoft, Google and Apple regularly send content with mixed-language content.

This is probably due to several factors:

1. Large web sites pull data for myriad sources, including CDNs and caching services. Each source needs to respect the requested language, If a source doesn’t have support for a requested language, then just that piece of content will be delivered in the fallback format.
2. Any proxies have to pass the requested language (and other headers) on to the backing server. If the backing server doesn’t get the language request, then it can’t respect the requested language, obviously.
3. Any proxy that caches content has to respect the language header (as well as any other data-relevant headers) instead of just caching one copy per URL. While this is standard for commercial proxies and CDNs, it might not be the case for bespoke software.
4. Some services might have a different context (e.g. logged-in user, detected via token in the request), who has different language settings than the requesting browser. This would mean that, while the main page content is pulled from the server with one language (e.g. en-US), the content for an embedded block might be requested as a logged-in user who has a different preferred language (e.g. de-CH). The server will likely honor the preferred language of the user account rather than the language included in the request, assuming it even even gets the language from the original request.
5. Finally, Some companies [1] are notoriously bad at multi-language software because they generally only acknowledge English and consider supporting other languages as a nice-to-have and that delivering English instead is an acceptable fallback because everyone can read English, right?

The move to cloud-based and highly cached content has increased complexity considerably. Even if a company does everything right in (1), (2), and (3) above, the realities of (4) may still lead to a page that contains content in multiple languages.

That is, each piece of software is functioning as designed but combining the output from those pieces of software leads to content that has multiple languages in it. At that point, you can either throw your hands in the air and give up…or you can start to redesign services to respect that requested language even if the user context’s preferred language is different. This is not a decision you can make lightly and you run the risk of breaking the service’s content in other places. Sometimes there is no right answer.

Since I live in Switzerland, which has 4 official languages, I’ve seen EULAs from Apple written in a combination of French, English, German and even a word or two of Italian.

The example below comes from Microsoft Edge’s Tips page that they show when you start using the browser. Edge thinks that my default language is German despite the fact that my Windows is English. Microsoft tends to use the language of the region you’re in (Switzerland) rather than the display language that you’ve expressly set, but…that’s another discussion.

At any rate, Edge thinks I want German content [2] but Microsoft can’t even reliably deliver German content for this main page, defaulting to English content in several places.

Microsoft Edge Tips Page

The Amazon Kindle Paper White is a pretty good deal if you’re already dependent on Amazon Kindle content. Kindle Unlimited and Amazon Prime are largely crap content with the occasional gem. If you’re already getting content from elsewhere, you’ll continue to do so because almost none of the content worth having is included in either of these services.

Replacing an E-Reader

I recently had to replace my old Amazon Voyage because I left it on a train and was unable to get it back. I tried reading with Kindle on the phone but, as nice as that app is, reading on the phone is no replacement for an E-Reader, at least for me.

I’ve had an E-Reader for almost a decade now. I started with the Sony Reader, which was good enough, but suffered from software and hardware deficits that finally put it to rest a few years later. About 4 years ago, I got an Amazon Voyage, with all the bells and whistles, and was quite happy with it.

Then I lost it.

So I read through the Guidemaster: Ars tests and picks the best e-readers for every budget by Valentina Palladino (Ars Technica) to get the lay of the E-Reader land. The two main contenders were the Kindle Paper White and the Kobo Clara HD.

I had actually used the mechanical buttons quite a bit, but these are apparently passé. Nothing but the top-end models of either the Kindle or the Kobo offer physical buttons. So I made my peace with it: no physical buttons.

The high-end versions are really nice, but they cost more than twice as much as a perfectly lovely reader. They have more LEDs for night-reading, more color ranges for before-bedtime reading. They have physical buttons. They are, by all rights, wonderful. Still, I decided it wasn’t worth the money for the slight perks.

So it was between a Kobo and another Kindle. The Kobo offered “clunky” Overdrive integration, which would have been nice to use with my NYPL subscription. I currently check out Kindle books and have to send them to my reader via Amazon. A direct integration and download would have been nice. I wasn’t looking forward to seeing what “clunky” means, though.

The support for EPUB books was also a plus, but the EPUB format from the public library is locked down with Adobe DRM, which likely would have been another uphill climb. Also, the Instapaper service only integrates with Kindles. I would have had to switch to Pocket for my weekly articles. I wasn’t really looking forward to switching that service as well.

On top of all of that, I actually have several Kindle books that I’ve recently been given or have purchased very cheaply. I would no longer be able to read those were I to switch to a Kobo.

I really wanted to get away from the Amazon hegemon, but I couldn’t bring myself to do it. The Kindle Voyage was one of the few high points in my otherwise fraught relationship with technology, software and hardware. It did everything I wanted—Instapaper on a regular schedule, library books via NYPL, notes in a neat and parsable TXT format—and held several weeks on a single charge. It never crashed. It occasionally rebooted. Even the UI kept getting better and better (especially highlighting). It just worked.

So I went for the Kindle Paper White.

Amazon Content Services

Lucky for me, it was Prime Day, so I got quite a deal on it—as long as I agreed to test-drive Amazon Prime for 30 days. Fine. €40 was a great discount from a reader that was better and only half the price of the Voyage I was replacing. On top of that, I found €28 on the German Amazon store, so I ended up spending very little to replace my lost Kindle.

So I signed up for Prime. In Germany. On top of that, I got access to Kindle Unlimited, but in the US. This is all a bit confusing, to be honest.

What is clear to me is that neither of these programs is worth paying money for. Unless something changes drastically, I will be canceling both before I have to start paying any subscription fees.

Almost every TV show I’d like to watch on Amazon is either not available in Europe or not included in Prime. Ditto for several movies I searched. Kindle Unlimited is not at all unlimited: not a single book I’ve searched for far—and I looked up about a dozen books from my wishlist—is included. A bunch of books I’ve already read are included, but that doesn’t help much—those are also available from the public library.

TV Shows and Movies on Prime

Amazon’s Prime offering in Germany for TV shows stands at 25% for the shows that I searched. I include them below.

I also searched for movies and the English-language offering is quite restricted as well. German-language movies I searched were not available.

The Expanse is not available for viewing.

The Expanse

American Gods is not available for viewing.

American Gods

Taboo is not available for viewing.

Taboo

Mr. Robot is not available for viewing.

Mr. Robot

Fleabag is not available for viewing.

Fleabag

Something called NOS4A2, starring Zachary Quinto, looks interesting, but also is not available for viewing.

NOS4A2

The Man in the High Castle is available.

The Man in the High Castle

The Marvelous Mrs. Maisel is available.

The Marvelous Mrs. Maisel

Books on Kindle Unlimited

Kindle books—even German ones—are only available from the American Amazon store.

Nation by Terry Pratchett is not included.

Nation by Terry Pratchett

Reaper Man by Terry Pratchett is not included.

Reaper Man by Terry Pratchett

I went through dozens of books on my wishlist; either the books aren’t available as Kindle books or they’re not included in “Kindle Unlimited”. At what point does the truth-in-advertising law kick in?

And the Kindle Unlimited Store is very sleazy. Here’s a “Gore Vidal” book that was written by someone else.

Gore Vidal 'book' that was not written by Gore Vidal

Here are a bunch of books I’ve already read. Cold comfort for me—and definitely not a reason to pay for unlimited access to books I don’t want to read.

Some books I've already read on Kindle Unlimited

Mostly, the Kindle Unlimited store is filled with flaky-sounding, generic junk by authors I’ve never heard of (probably most are self-published).

A bunch of junk by people I've never heard of

And then there’s the obligatory giant pile of softcore, light BDSM, husband-material, soloflexed romance novels. The NYPL is also littered with this tripe. Both services’ pinheaded recommendation algorithms love this content and suggest it for anyone who just searched “Chomsky” or “Lem”. It makes you feel a bit more secure that the AIs aren’t quite ready to take over yet.

A giant pile of softcore romance 'novels'

To wrap things up, the Amazon UI also makes it very easy to think that maybe you’re just not properly signed up for “Kindle Unlimited”. I clicked the big link at the top because I thought I was either (A) on the wrong site again (e.g. Amazon.de, where the service doesn’t exist) and using an account where it wasn’t activated or (B) had to jump through yet another hoop to activate it.

You have to read the tiny text at the end carefully to see that I have Kindle Unlimited, but that the offer they’re making is to “add 6 months to your existing membership”. It would be very easy to slip up and order 6 more months in addition to the 3 free months of a useless service, in the mistaken hope that you’d finally see some interesting books.

Sleazy attempt to get me to sign up for 3 months o… Unlimited before my free trial is even 5 days old

Conclusion

It seems the Amazon’s giant warehouse of content is filled with crap. Anything worth watching or reading still costs money. If you subscribe to these Amazon services, then you’ll likely start gravitating to watching or reading crap just because it’s included in your subscription.

If, however, you like to choose what you watch or read, neither of these services will save you much money. You’ll still have to shell out to get most of the content that you’re really interested in.

Disclaimer: these are notes I took while reading this book. They include citations I found interesting or enlightening or particularly well-written. In some cases, I’ve pointed out which of these applies to which citation; in others, I have not. Any benefit you gain from reading these notes is purely incidental to the purpose they serve in reminding me of what I once read. Please see Wikipedia for a summary if I’ve failed to provide one sufficient for your purposes. If my notes serve to trigger an interest in this book, then I’m happy for you.

This paper (arXiv) focuses only on intra-process attacks, advocating for mitigation via separation into multiple processes, using hardware-level protection that provides a stronger guarantee.

“This paper explores speculative side-channel attacks and their implications for programming languages. These attacks leak information through micro-architectural side-channels which we show are not mere bugs, but in fact lie at the foundation of optimization.”

This paper proves unequivocally that software mitigation of intra-process, side-channel attacks are futile. An attack is always possible, with a combination of scaling for any-resolution timers and patience. “As we have seen, access to a timer, no matter the resolution, leaks µ-architectural information.” Type-checks in any languages may partially mitigate, but this paper proves that they cannot full mitigate and are, therefore, largely useless for data-security.

The only difference is the bit-rate of extracted information, not whether information can be extracted. They not only proved this with theorems, they implemented many of the attacks to provide accurate estimates of the expected extraction bit-rates in the presence of various mitigations.

There is one particularly virulent variant (#4) that has literally no mitigation.

“Mitigating type confusion for stack slots alone would have required a complete redesign of the backend of the optimizing compiler, perhaps man years of work, without a guarantee of completeness.”

As anyone who’s been following this problem suspected (or pretty much already knew), the world traded of security for performance long ago. Though chip manufacturers and operating-system designers paid lip service to security, performance improvements were paramount.

We’ve known since the end of 2017, but now we have proof. The paper sums it up more nicely than I could,

“Our models, our mental models, are wrong; we have been trading security for performance and complexity all along and didn’t know it. It is now a painful irony that today, defense requires even more complexity with software mitigations, most of which we know to be incomplete. And complexity makes these three open problems all that much harder. Spectre is perhaps, too appropriately named, as it seems destined to haunt us for a long time.”

Citations

“This paper explores speculative side-channel attacks and their implications for programming languages. These attacks leak information through micro-architectural side-channels which we show are not mere bugs, but in fact lie at the foundation of optimization.”

“We were able to leak over 1KB/s from variant 1 gadgets in C++ using rdtsc with 99.99% accuracy and over 10B/s from JavaScript using a low resolution timer. We demonstrated a potential 2.5KB/s variant 4 vulnerability, but with low reliability, starting at 0.01% but amplifiable up to 20% through various techniques. We found that using shared memory to construct a timer worked well enough in JavaScript to measure individual cache hits and misses and exploit any of the known leaks.”

There are many places where you can authenticate on MacOS. I think the following list is comprehensive.

• MacOS
• App Store
• Messages
• FaceTime
• iCloud
• iTunes

Each of these logins can actually be different, believe it or not. Most of the time, they’re not, though. Apple prefers if you just use the same account for everything.

If you, for historical reasons—purely hypothetically, understand—have more than one account, you might log in with inconsistent accounts.

Normally, it doesn’t really matter. That is, unless you log in to iTunes with user A when user B has all of the music purchases. Or, vice versa, if user B has all of the App Store purchases.

I had the following logins on my work laptop:

• MacOS: local account
• iCloud: marco1
• iTunes: marco1
• App Store: encodo
• Messages: marco1
• FaceTime: marco2

The FaceTime login was an “error”, left over from before I’d consolidated my accounts. I only discovered this after I’d solved the problem I had.

Since it’s a work laptop, I use “encodo” for the apps and my own account to pull in everything else on my user account.

My problem was that I was no longer able to update software installed with user “encodo” because Apple kept asking me for the password for “marco2” in addition to the password for “encodo”.

App Store Uneditable Apple ID

At the time, I had no idea why the App Store was asking for this login. I couldn’t figure out why it needed a second login, in addition to the one with which I’d logged in to the App Store. On top of that, the username was read-only, so I couldn’t even tell the App Store to use the “encodo” user or “marco1” instead.

It was only after I’d applied the latest system update and restarted that I learned where that login had been used. On restart, two things happened:

• MacOS asked me to log in to iCloud during setup. That’s odd. I’m already set up to automatically log in to iCloud for a couple of things. Why does it ask for the password? At any rate, I don’t have the password at this point because I don’t have access to my password manager yet. I ignored it and continued setup and initial login. Once I’d logged in, I checked System Preferences. I was logged in to iCloud, as I’d assumed. Weird.
• MacOS then asked me to log in to FaceTime with “marco2”. Aha. The penny dropped and I began to suspect more strongly that my failed app update was related to Apple’s not having my FaceTime password stored. That is embarrassing for Apple, I think. I changed the account to “marco” and continued.

After I was back in, I returned to the App Store. This time, the update of the app went through without a hitch. Thanks, Apple. Maybe we can get logins sorted out by 2020?

I saw in the news today that a new version of the Alpine Linux distro is available. It made me wonder whether those who work on distros like Alpine—that a few years ago would have been more rarely used—are excited at the uptake in usage for containers.

Working on open-source software has always had a bit of altruism bound up with it: if you don’t get compensated—most interpret this as remunerated—then why do it? There are a host of reasons to work on something, besides getting paid for it. Many get satisfaction simply from doing the work, and/or knowing that others are able to benefit from it.

This logic always seemed applicable for smaller software—which doesn’t require a large commitment on the part of the user to install. In order to get someone to use an operating system, though, a user has to make a much larger commitment.

Until recently, someone could install an open-source distro side-by-side with their “main” operating system, but they would still end up using it less. There are many of us who would have loved to use Linux on more than just on servers, but couldn’t move away from Windows entirely. VMs were helpful, but still not a good solution.

At least that used to be the case. Nowadays, we have containers instead of VMs and we’re all using Linux more than ever. In fact, I’m used to using Debian or Linux for servers, but have been using something called Alpine Linux for my Docker containers that only need to host a PostgreSql or SQL Server or build and run some .NET Core tests.

It’s nice to think that the increased prevalence of containers is increasing usage of smaller distributions, making it ever-more likely that open-source work will be rewarded by increased use.

tl;dr: If your birthday-calendar appointments don’t show up on your MacOS iCal, try turning it off and on again (rebooting the machine). That’s what eventually worked for me. For the full story and more tips, read on.

A little while back, I enabled iCloud synchronization for my contacts. That didn’t go so smoothly on the first try: the MacOS desktop that had most of the information was actually erased because Apple doesn’t seem to understand what “sync” means.

Using a backup, I was able to save my contacts and get my three devices on the same page. It’s quite nice that the contact changes are automatically disseminated to three devices at once. I used to have syncing for the desktop and the phone, but my other laptop was not privy.

All in all, despite the initial stumble, it seemed to be working fine.

Syncing Calendars

Somewhat encouraged, but with misgivings, I thought maybe I could synchronize my calendar in the same way. I’ve started using the calendar more, but don’t sync to my phone often enough. The cloud could be quite useful here, too.

Fingers crossed, I enabled iCloud synchronization from my desktop (where my calendars are).

I turned on iCloud calendar-sync on the phone, but didn’t see any immediate difference since the calendars were already synced. I didn’t see anything happening—of course, it’s Apple, so progress bars or circles are superfluous and disturbing—so I went off and did something else.

A bit later, I wanted to check my task list, which I keep in Reminders.

All of my lists and reminders are gone.

I mean, obviously.

Reminders and iCal are conjoined twins

I hadn’t enabled Reminders in iCloud. Why did it wipe those out?

Reminders and Calendar are basically inseparable. Apple lists them as separate sync targets in the iCloud settings, but I don’t believe that it’s actually possible to sync one without the other. The data is all mixed up in the same folder in your Application Support area.

I opened iCal. All of my calendars were gone, too.

Coooool.

Incompetence

After my recent experience with Contacts, this no longer came as a surprise to me, so I spent a bit less time throwing things around my office and swearing at the top of my lungs about the utter incompetence of a company absolutely suffused with the best engineers money can buy that can’t even get a simple synchronization right.

It wasn’t even a synchronization—it was a merge of a list with an empty list. And their brilliant sync-process ended up with an empty list without asking a single question.

What the actual fuck.

Basically, the same bullshit that happened with Contacts.

It’s made worse by the fact that Apple is so confident of the flawlessness of their software that they provide no logging, no monitoring, no introspection and no UI or help to address problems.

It’s just broken or not working or sporadically working or your data is just gone and there are no switches or toggles to adjust.

The Long, Scary Road to iCal Syncing

First I turned off iCloud sync for iCal.

Then, I turned immediately to the same solution as with contacts: restore from backup.

Easier said than done; the iCal backup is harder to find than Contacts.

Also, restoring the “Calendars” folder doesn’t work. My calendar and reminders were still empty.

I can see all of the individual files with the correct data, but neither iCal nor Reminders reads it or shows it to me. [1]

I logged in to iCloud and saw all of my calendars, as expected. What’s going on? It looks like iCloud uploaded my data, then deleted it from the source, put on its hat and went home for the weekend, patting itself on the back for a job well-done.

This can’t be the state of engineering for such basic software, can it?

Oh, but it is.

On a hunch, I restored the version that iCloud has left me and enabled iCloud synchronization not only for calendars but also for Reminders. In seconds, my calendar and reminders were, once again, populated.

Kind of.

iCloud and Mac had managed to paste together a mish-mash of long-forgotten calendars and reminders lists, complete with duplicate appointments and reminders.

Also, the birthday calendar was enabled, but none of its appointments were visible.

Turning it off and on again did nothing.

Putting it all back together

I enabled the birthday calendar in iCloud. It informed me that it could take up to an hour to calculate the birthdays—from my 150 contacts. Fine. That sounds reasonable.

The birthdays finally showed up in iCloud, but still not on my Mac. The iPhone showed them just fine, along with the plethora of calendars that iCloud had put together for me.

I deleted some of these calendars and reminders and could see them dropping out of iCloud. I changed the colors of some of the standard auto-generated calendars to de-emphasize them. At this point, I was too terrified to delete any of Apple’s calendars for fear that their algorithms would punish me. At each step, I was reassured to see iCloud updating more-or-less in real-time.

I checked my calendar on my phone and saw that it hadn’t synced yet—there were still all of these other calendars displayed. After a minute or two, they still hadn’t updated on my phone. The phone had even more auto-generated calendars than the desktop.

I turned with dread to my desktop to see that, once again, Apple had applied its inexorable and inscrutable “merging” logic to revert all of my previous changes and more-or-less use only the versions it had found on the phone.

It was incredible, but it looked like Apple’s merge policy was to apply a sane policy for two sources that had been introduced but, for a new source seemed to just let the stranger barge its way in and overwrite everything that had existed before.

I know that that can’t be the way that it works, but that has been the sum-total of my experience so far.

Oh my God are we synced yet?

I changed some colors back and could see that both iCloud and the phone were now updating, exactly as expected according to the logic I’ve outlined above.

I still didn’t have birthdays on the desktop, though.

In a final, desperate move—but without much hope—I rebooted the desktop. It’s so quick these days: rebooting, logging in and having MacOS restart everything takes only a handful of seconds.

Aaaannnnddd….the birthday calendar is working again.

Incredible. And pathetic.

Also, I have all of my old calendars back. Again. With original colors. Again.

On top of that, now I’ve got two copies of the “US Holidays” and “Swiss Holidays” calendar: one set in the “iCloud” area and one in the “Other” area. I check iCloud: it doesn’t have these two calendars. MacOS swears that those calendars are in iCloud and iCloud tells MacOS to “go home, you’re drunk”. My phone agrees with the desktop. Fun times.

Seriously, this is the last step

So, which of the duplicated calendars should I delete? The ones in “Other”? If I delete those, will iCloud at some point sync and remove the copies from my desktop and iPhone, leaving me with no holiday calendars? Who knows?

Duplicated Calendars

One more time, with feeling. Delete local calendars, leave Apple auto-generated calendars alone, unsubscribe “Other” calendars. What to do with the local “Trips” calendar? I exported it, then imported it to a new calendar, which automatically created the new calendar in iCloud. Delete the original from “On My Mac” and Bob’s your uncle.

Finally, everything seems to be in-sync and all calendars and reminders are working on all devices.

Maps

Later that morning, I was alerted to an appointment in the afternoon. I’d attached the contact/address, so the alert showed me the map.

Nice integration. Let’s use it.

Click.

Maps opens and shows me the location. I get directions. Hey! They have public transit for Switzerland now! I don’t have to use Google Maps for this stuff anymore? Cooool.

I’d looked it up on the desktop—let’s share it to the phone, shall we? There’s a little “share” button everywhere in MacOS. There’s even an entry that lets you “share with iPhone”.

Click.

Nothing.

Click.

Nothing.

Share with Messages. Send to me and my wife.

A message pops up with a map that I’d looked up 2 weeks ago. On both phones.

I tried again, to no avail. At best, I’d get a message with the address of a restaurant we’d eaten at 2 weeks ago.

Fine. I looked up the address on my phone and got it that way. It didn’t take long. That’s not the issue. The issue is a company with so much talent, data and software getting so close to useful integration and then dropping the ball on the one-yard line.

Who the fuck is writing this amateurish and terrible software? It’s an absolute scandal that this is the state of such essential and simple software components in 2019.

The integration is so promising and then stumbles on the simplest of cracks in the road.

Will Artificial Intelligence take over the world? Only if we let it, I think.

And we’re almost certainly going to let it.

Our habit seems to be to capitulate to any form of power that dangles a short-term bauble of convenience before our greedy, beady eyes, even if it always seems to be just out of our grubby reach.

AI is dangerous less for what it is capable of doing now and more for how much power we concede it even when it’s stupefyingly shallow. We’ve given up before the battle has even been joined, conceding vast swaths of the intellectual landscape to an enemy that hasn’t even proven itself close to worthy of our respect.

That’s why we’re perennial losers. We don’t even respect ourselves.

Can you imagine how much power the AIs will have once they get really intelligent, once they stop messing about with making us buy things and think of more sinister purposes to which they could put their organic army?

We’re already more-or-less enslaved to our reasonably shiny and relatively advanced devices, offering the primitive content and experiences that they do. We have a complete paucity of vision and we’ve passed that on to our electronic offspring.

I’m not too worried about the AIs taking over what I do. I’m completely expecting everyone else’s capitulation to AIs to make the world we live in an even sadder place, flattened to a search-bubble homogeneity and bereft of risk or fun.

They’ve already gone a way toward doing so: wherever the algorithms have impinged themselves, they suck our time and creativity and pull us into a world just as programmed as our own.

How many decisions do you even make on your own these days anyway?

Robot armies striding the post-apocalyptic hellscape with conventional weapons trained on any furtive, mammalian movement seems like a very predictable and manageable end-game compared to the amorphous, digitized blob that is modern, Western life.

From Kai-Fu Lee’s new book says Artificial Intelligence will be Google vs China and will kill half the world’s jobs by Robert X. Cringely

“And I find it difficult to see China as being any more of a development crucible than is Sand Hill Road, where startups have even less time to succeed and therefore more pressure to evolve. Explain to me how this is incorrect, because the numbers are published and are real. In comparison to other adolescent startup cultures, yes China is and will be successful and they are (finally) getting it, but Silicon Valley still has more of everything including investor impatience.”

I fucking hate everything about this mentality. It forsakes the good for that which is the bare Minimum required to make money. I call bullshit on your supposed innovation, because it’s at best whale vomit, a nominally useful byproduct of a process with a completely different goal.

The goal is to hoard wealth.

the inefficiency of the process whereby we also incidentally get something useful is a luxury we can no longer afford with the resources and capacity remaining on this planet.

If we don’t it, it’s unlikely our AIs will respect it until they get smarter than us, by which time it will be too late. But go aead and fiddle, shuffle your deck chairs. Who can stop you?

“And very rarely my partner would be called-in just in case I could find in Silicon Valley some garage in which time travel had already been conquered. I was usually able to come through and occasionally the big company even came out of the experience looking brilliant.”

What’s your point Bob? That you’re brilliant and only Americans are capable if winning a game they’ve rigged? You think it’s fair? That it’s a level playing field? What an ass. Congratulations, you can chew through the planet more indiscriminately than everyone else.

“The China environment today clearly does not have the level of integrity found in the US, but it has gone a long way.”

WTF. China is supposed to be striving to have the integrity of the States? That’s a goal? After LIBOR, after 2008, after the last 100 years, how can anyone think that U.S. business has any integrity?

“As an American teenage male in the 1960s, having or getting a car was the single most important life passage. Not anymore. That’s an enormous change that lowers income requirements and makes it easier to function in an AI-inspired culture of economic limits. (Emphasis added.)”

That’s the conclusion you draw? Also, nice euphemism (highlighted).

“But what I CAN predict is that we won’t go into a dystopian future because my kids wouldn’t stand for that. There is a huge correction coming and the only thing my kids believe is that they will control that correction.”

Looking to those with a surfeit of unearned confidence for solutions smacks of desperation.

This is a laundry list of issues I’ve had with iOS over the last half-year or so. Some things get fixed; others break. This operation system is in its 12th version and is 10 years old. It’s made by the richest company on the planet. It’s frustrating to watch the magical engine of capitalism and privatization be seemingly unable to get even the easy stuff right.

How is it that I’m the only one to whom this happens? iOS is supremely unreliable. I have an iPhone 6s with iOS 12.01 and a battery with 94% health.

Location services

They’ve TWICE NOW been mysteriously back on. System upgrade? Stop touching my stuff, Apple.

Messages

• Messages are duplicated
• Messages don’t go out, but aren’t marked as undelivered
• Some conversations suddenly turn green (SMS-only)

Power Draining

iOS still doesn’t have anything to prevent apps from misbehaving

• Clock has sucked away 60% of my battery, driving it into the ground
• So has Garmin Connect, with background activity

Clock: 10 minutes on-screenClock: 60% of batteryGarmin Connect: 59% of battery in background

• I can’t block background activity; I have to kill the app.
• iOS doesn’t allow you to set a power-usage limit for apps.
• Power-save mode must be approved: it can’t just turn on automatically under a certain level
• So, at work, my phone just happily drives itself into the ground, making sure there’s no battery left when I need it

Bluetooth

It won’t stay off. If you’re in airplane mode, it shows up as “off” (disabled), but it’s really enabled and temporarily disabled. Once you’re out of airplane mode, it comes back

Camera

My camera has also crapped out so badly that restarting doesn’t help. I had to reboot.

Personal Hotspot

The Personal Hotspot comes on every time I plug my phone in to charge. I can only turn it off by putting the phone in flight mode. When I turn it off in the settings, it seems to turn off, removing the blue bar at the top, but … when I go back to the settings, it’s back on again

Mail

The “cellular data” toggle in the Mail settings has the same issue: you can toggle it off, but it’s still on when you go back into the settings. The only way to actual toggle off data for Mail is to do it from the cellular-data settings. The toggle in the Mail app is just decoration: it looks like you can change the setting, but the toggle is actually read-only.

Mail has had such power-draining background activity (no logs or indication why, of course), that I’ve turned it off entirely.

Bonus: MacOS ICloud

I finally cracked and turned on a trickle of data (PhotoStream) in iCloud.

I was unable to log in at first because I was presented with a model dialog with no fields or buttons.

iCloud Login

For those hoping for help, I honestly can no longer remember how I got around this. Probably restarted the app or disable iCloud and re-enabled it.

Here is the text of my bug report to the SBB for its iOS App. I wrote them in German, so I’ll let them stand that way; translations are in the footnotes.

Verbesserungsvorschlag im iOS-App

Umgebung

iOS 11.4.1, SBB Mobile 8.2.2 (10)

Use Case

Einen Benutzer will schon geladenen Streckeninformationen ohne Datenverbindung lesen.

Beobachtetes Verhalten

Der SBB-App aktualisiert immer wieder automatisch, auch wenn der App nur vor kurzem verlassen wurde (z.B. 10 Sekunden, wenn man auf einem anderen App für Infos gewechselt hat). Der App löscht vorerst alle wichtige Daten und danach versucht “bessere, neuere” Daten herunterzuladen. Wenn das nicht gelingt, bleibt der Benutzer ganz ohne Informationen. Einen Benutzer verzweifelt an dieses Verhalten besonders, wenn ganz offensichtlich keine Datenverbindung entsteht (z.B. Flugmodus, um Daten/Akku zu sparen) und die nötige Daten nur flüchtig auf dem Bildschirm gelesen werden könnten vor dem die mit einem “Oops. Something went wrong.” ersetzt werden.

Erwartetes Verhalten

Der SBB-App aktualisiert Infos über eine Fahrtstrecke nur automatisch, wenn eine Datenverbindung entsteht. Die Daten auf dem Bildschirm, auch wenn die nicht top-aktuell sind, werden erst nach dem Erhalten von neuen Daten gelöscht und mit aktuelleren Daten ersetzt. Zudem aktualisiert der SBB-App nicht immer, wenn der App wieder Fokus hat sondern nur nach einer bestimmten Zeit (nachdem man erwarten könnte, dass neuere Daten überhaupt vorhanden sind … diese Frist ist länger als eine Minute … eher fünf). [1]

They wrote back a couple of weeks later:

“Gemäss Rückmeldung unserer Fachspezialisten ist das Verhalten in der Zwischenzeit bekannt und es sind Optimierungen geplant.

“Wann die Umsetzung erfolgt ist noch nicht bekannt. [2]”

This was several months ago. I haven’t seen the fix yet—as of a couple of weeks ago, the app was still clearing data.

Windows Search has been unpredictable for a long time. If you’re a MacOS user, it feels terrible. It’s an utter mystery how Windows can’t seem to find anything, even in a small pile of startup icons and control-panel entries. It’s a database of a few hundred entries, at best. Let’s see how Windows 10 fares.

I finally took the time to document my struggles to run the “Microsoft SQL Server Management Studio”. There are so many words in there. Which ones can I type to find the icon? Hint: I should be able to type a piece of any of the words to get that icon. Spoiler: I cannot.

Type ‘sql’

Let’s type SQL and see what happens.

Type 'sql'

It finds SQL tools that I rarely or never use, but not the one I use all the time (and that I’m searching for).

Type ‘sql mana’

Ok, I’ll add “mana” to the search phrase to try to force Windows to match the icon with the word “manager” in it.

Type 'sql mana'

This matches stuff that looks right, but its only folders, not the application icon.

Type ‘sql s’

The word following “SQL” in the icon is “server”, let’s try to trigger that.

Type 'sql s'

Everything but the actual program I’d installed.

Type ‘sql se’

Let’s keep going.

Type 'sql se'

Oooo, getting warmer … but still only folders.

Type ‘sql serve’

Type 'sql serve'

Finally, after typing most of the name, I get the icon I was looking for. It’s unclear why it matched better now—all of the letters in the search phrase are in all of the other icons, as well.

Type ‘st’

That gave me an idea: how about I type letters that show up only in that icon?

Type 'st'

Nice, Windows: You found the app with the word “Studio” in it that I use the least.

Type ‘stu’

Type 'stu'

Ok, with one more letter, you’re getting warmer. I use Visual Studio 2017 the most. And I see that the Management Studio has shown up on the radar.

Type ‘m’

Finally, I tried typing the first letter of the first word of the application name. I got a match almost immediately.

Type 'm'

Type ‘ma’

If I add the second letter, then I get a hit with just two letters. It’s honestly a mystery why this is a hit whereas the other two-letter combinations are not.

Type 'ma'

In conclusion, here we are with the third or fourth version of Windows 10, the newest of a line of operating systems that stretches back 3 decades. It’s 2019 and just finding an application to execute—and one made by the same company that made the operating system—is a heroic undertaking.

Use Case

Let’s imagine we’re working on a PHP web site together, using PostgreSql as a database.

Without Docker

Without something like Docker, I’ll write a readme.md that tells you which PostgreSql to install (maybe latest, whatever), how to configure the Apache server (or Nginx, whatever) and make sure the document root, extensions, modules, etc. are all lined up for this project.

In order to write this readme, I had to set it up on my machine and carefully write down instructions matching what I did, leaving nothing out.

I probably won’t get it right. Or maybe one of those settings conflicts with settings required by other projects I’m working on. Maybe I get lost in a configuration/setup rabbit hole for days. Maybe you do.

Your local development environment can get pretty overloaded with dependencies and configuration, some of it for projects you’re no longer even working on. A hack you made to your environment for one project may subtly affect other projects.

Use a VM?

What’s a solution? Run all of the stuff you need for your project in a VM. One VM per project.

That’s a lot of work and maintenance, too. We either have to pass around giant VM images every time one of us makes an update or keep updating that readme.md and apply the changes ourselves.

Scripting to the rescue!

But what if we could turn that readme.md into a script that you just run to create the environment needed for that project? Instead of maintaining the software, you just run the script and it installs and launches everything you need (web server, extensions, database, file-mappings, etc.) on the fly?

That’s basically what Docker does.

Instead of writing instructions for you to execute, I write instructions for Docker to execute. Instead of hoping that my instructions also work for you, I’m running the exact same script you are. If you add new functionality (e.g. that depends on a new PHP extension), your pull request includes not only your software changes, but the configuration changes for the environment to ensure that I also have that extension enabled when I test or run your code.

When to use Docker?

You work alone? Doesn’t matter. It’s still better to work this way because you’re not messing up your machine for each project. Also, what if you need to work on a different machine? What if your drive dies and you need to restore? All of this is super-easy with a script-based environment like Docker.

Final Example

I have a PHP project that I only very occasionally work on. It used to be a nightmare to get everything set up and running again after 6 months or a year, just for a small bug fix. WHY ISN’T PHP-DEBUGGING WORKING?

Now? Now I have a Docker compose file that pulls up a database server, a web server with all required extensions and a content container. Debugging is configured right in the scripts, so I never have to fight with that again. When I need to work on this project, I just docker-compose up and I’m off and running. Anyone who downloads my code can do the same thing. [1]

Apple iCal is another piece of software that’s in a very established field, with a very established feature set, in which Apple has been producing software for over a decade. It’s a calendar application with reminders. The reminders can be set to a specific time, with one or more alerts. An alert can be snoozed for a certain amount of time.

This is not rocket science.

Ok, so a modern calendar has to be able to pull in remote sources, to sync with other sources, and to send notifications. It manages multiple calendar sources and lets you choose the color of calendars and appointments. There’s also a merge function to move entries from one calendar to another.

I still don’t see any rocket science here.

At least, not for a company with over 10,000 engineers. [1]

So, after a dozen versions of a piece of software in a simple problem domain produced by one of the largest and richest and smartest engineering firms the planet has ever seen,

• Why is it that the reminder apparatus is still so primitive?
• Why are the defaults to just keep reminding you every five minutes until you answer its question?
• Why can’t you adjust how snoozing works?
• How is it that Apple isn’t even capable of detecting when you’re running a full-screen app/movie and not show its alerts over top of everything?
• Why so pushy? You have a notification center, for God’s sakes.

I’d like to be able to auto-snooze after three reminders instead of hearing that infuriating ping every five minutes until I get up and do what the computer wants.

The computer is not in charge. This is something we’ve forgotten. [2]

Humans are currently better at detecting patterns than machines—this may change in the near future. [1]

However, machines are massively better at sorting detected patterns than humans. Humans can’t stick to a regime.

For example, a site like Reddit amounts to a vast sorting algorithm with posts, pictures, videos and articles as input and the subreddits as buckets.

Myriad users upvote and downvote these items to determine whether an item belongs in a given subreddit. If an item in one subreddit would be appropriate in another subreddit, a user can cross-post it to the other subreddit.

But humans suck at this.

As a specific example, look at a forum like /r/oddlysatisfying. The material you should expect to find there is in the name: images or videos of everyday objects that provide a soothing experience.

“This can be physical (like popping bubble wrap), visual (a perfectly looped GIF), or even aural (the crunching of leaves). […] subjective. Indefinable. Completely recognizable. It’s a wonderful feeling.”

For example:

Coffee-cup Pyramid

Something like Casual Japanese Wrestling absolutely does not belong there. It just shows a neat backflip in a professional wrestling match.

Two challenges arise when using humans to categorize:

1. In order to get humans to participate, there has to be a reward system. It’s either money or it’s something else. That something else is called karma on Reddit. Humans aren’t just happy to get the karma appropriate to their behavior; instead, humans modify their behavior to get more karma. That is, instead of sorting data in the system, they let the system train them to sort data. This leads to rampant cross-posting, where people take material that garnered a lot of upvotes in one subreddit and re-post that material in other subreddits to “farm karma”.
2. Most people don’t really pay attention to which subreddit they’re browsing. They read the title, view the image or watch a few seconds of video and then decide it’s “cool” (upvote) or “stupid” (downvote). They don’t generally think about whether the item is appropriate to the subreddit. That is, you may think it’s cool, but is it oddly satisfying? If not, then you should downvote it in that subreddit.

Taken together, these behaviors lead to a tremendous amount of content being duplicated across a plethora of false categorizations.

Where a machine can be trusted to stick to its rules, applying them rigorously, humans are distracted by shiny objects and make for poor sorters.

tl;dr: If you don’t use your voice mail for your phone, then you should disable it. It is ridiculously insecure and can or will be used by hackers to gain access to other services you use (e.g. Whatsapp or PayPal).

On Sunrise, you can deactivate your voice mail by “calling” #145#. To re-activate, call *145#. You’ll get a confirmation message.

The CCC annual conference (Chaos Computer Club) took place last weekend, in Leipzig, Germany. There were several interesting talks, but this one stood out: Vigo’s presentation was very good and didn’t seem as padded as other presentations. On top of that, it’s a real-life hack that can affect anyone with a phone and active voice-mail.

What does he do?

He takes us through a bit of history about hacking old phone systems—the “phreakers” of the good, old days would hack systems with special sounds (DTMF tones) or by flooding poorly written checking algorithms with more data than they was expecting. Using these tricks, they would gain access to free international calling or be able to charge calls to other people’s accounts.

It still works today.

Wait, how? We all have Smartphones now. We barely even use voice-calling! Nobody uses voice mail anymore.

All very true. But: just because you’re not using your voice mail doesn’t mean that someone else can’t use it.

His hack is multi-stage. I’ve summarized a bit below, but you should watch the video for the exact steps. [1]

1. Find a phone number that user has set up as a 2FA for the service you want to crack (he shows examples with Whatsapp and PayPal)
2. Get the service type for that number (e.g. O2 or Vodaphone)
3. Call the service number for that phone’s voice mail
4. Brute-force the PIN like it’s 1975 (takes a few seconds)
5. You now have access to that number’s voice mail
6. Make sure that phone’s calls will go to voice mail (several techniques here)
7. Go to the service and request a passwood reset, but say you want them to call you instead of sending a text
8. The service calls with an automated message
9. It lands in voice mail
10. Use your software to load the voice mail and send it through voice-recognition software to get the code it read aloud
11. Record this code as DTMF in your outgoing message
12. When the service calls again to verify the code (you’re supposed to type it in at that point), your outgoing message will play the DTMF tones at the correct time and the service thinks you typed it
13. You can now proceed with password-reset on that account because you’ve “confirmed” that it’s your phone, when all you have access to is the voice mail.

35C3 − Compromising online accounts by cracking voicemail systems by Martin Vigo (YouTube)

It’s easy enough to make fake text and fake articles. Anyone can write something online. You can make up information out of whole cloth or you can subtly manipulate the so-called truth to slant it in a desired direction, either by omission or by emphasis or both.

Since many people read only the titles of articles, you don’t even need to try too hard to make well-substantiated and well-linked articles. Sure, somebody’s going to “debunk” your text. However, by then, your lie will have made its way around the world and established itself as truth in many minds. The debunking will either be ignored or actively repudiated. You will have established mind-share with your lies

You can lend gravitas to your article with quotes and citations. Even more effective are videos and images. People trust videos and images more than text. That’s why you’ll often see stock photos that have nothing to do with the article. Or the image is related, but is of a different place or from a different time. If you’re writing an article about a natural disaster but the pictures you have available aren’t sufficiently tragic, then you’ll just pluck a stock photo and trust that no-one notices that it’s of the wrong island or from 10 years ago.

This works for people, too. You can lend credence to your fake quotes by just including a photo of the person who supposedly said it. As shown below, now you can include a video.

Fake videos of real people – and how to spot them by Supasorn Suwajanakorn

This video is from April 2018. The presenter describes software that can build videos depicting people saying things that they never said. It builds fake videos. This software is real. Almost no-one is adequately equipped to avoid being duped by such videos, at least temporarily.

He goes on to indicate that we can detect fakes—at least for now—because the technology for properly emulating teeth and tongues is about ten years away. This is cold comfort. As is the browser plugin that he says he is also developing, which will detect these kinds of videos in real time.

What is far more likely is that fake videos like this will sweep around the world—and perhaps already have. They will sow discord and unrest until something really, really bad happens. Imagine a video of Putin declaring war on the U.S.

If we could be manipulated with text and pictures, imagine what can be done with this video technology.

In another, similar development, a team at Nvidia has developed new technology that can create completely new images from existing images, by analyzing and mixing similar images. The paper A Style-Based Generator Architecture for Generative Adversarial Networks by Tero Karras, Samuli Laine, Timo Aila (arXiv.org) is online.

The demonstration below is quite impressive, showing completely artificial human faces that are completely convincing. They move on to objects, like furniture and entire interior designs as well as cars. It’s quite impressive how smoothly it mixes in real time.

A Style-Based Generator Architecture for Generative Adversarial Networks by Tero Karras FI

This technology, too, could be used to create extremely convincing and false pictures to decorate false news reports. A picture with the proper characteristics lends a large amount of credibility to a story, at least initially.

An interesting place to use this technology would be in police investigations, to replace sketch artists (if they haven’t been largely replaced by similar technology already). The sketch artist could be trained to mix the proper characteristics to create a convincing face until the victim or witness has a eureka moment and recognizes the “photo”.

Update (31.12.2018): even without this face-mapping software, it’s possible to manipulate videos with desktop software, as shown in the clip below, from the Neo Magazin Royale from ARD Fernsehen in Germany. In it, they made a fake video of Yanis Varoufakis giving Germany the finger. It was convincing enough that the largest news-talk show featured the video (without checking its origin) and then had Varoufakis on the show to explain himself. Hilarious, but definitely a sign of the future. What if no-one had admitted they’d faked it?

The video is in German with English subtitles.

—Jan Böhmermann (YouTube)

Did you know that there are some filenames you’re not allowed to use in Windows?

I knew that there are characters that you cannot use in a path, like : and ? and *, but I didn’t know that otherwise innocuous-looking names like aux.txt and con.txt are verboten.

I learned about this from the article It is 2018 and this error message is a mistake from 1974. This limitation, which is still found in the very latest Windows 10, dates back to […] by foone (Thread Reader), which has a pretty click-bait–y title and is quite long, to boot. The author includes a lot of detail and conjecture about how it is that Windows 10 still has some forbidden/reserved filenames (and then takes half of it back or corrects it in several long footnotes).

Instead of citing the article, I included the upshot that I found interesting, which is better explained in the official Microsoft documentation Naming Files, Paths, and Namespaces (Windows Dev Center), referenced from the article above

“Do not use the following reserved names for the name of a file:

“CON, PRN, AUX, NUL, COM1, COM2, COM3, COM4, COM5, COM6, COM7, COM8, COM9, LPT1, LPT2, LPT3, LPT4, LPT5, LPT6, LPT7, LPT8, and LPT9. Also avoid these names followed immediately by an extension; for example, NUL.txt is not recommended.”

While it’s possible that this is only a recommendation, Explorer.exe does not allow these filenames at all, as shown in the screenshot below (take on an up-to-date Windows 10 Enterprise installation).

I tried nul and con with several different extensions (bmp, exe, txt, doc, etc.) and none of them are allowed. I have not checked whether this restriction applies to files created programmatically (i.e. is the restriction enforced by Explorer or by CreateFileEx?)

I’m just wondering how I’ve never noticed or heard of this before. Talk about a restriction that Microsoft was wise never to fix because people don’t run into it! I guess the filenames are so short I’ve never tried to use them.

From the article New botnet infects cryptocurrency mining computers, replaces wallet address by Dan Goodin (Ars Technica)

“Records show that the attacker-controlled wallet has already cashed out slightly more than 1 Etherium coin. [1] The coin was valued at as much as $1,300 when the transaction was made. At the time this post was being prepared, the records also showed that the attacker had a current balance of slightly more than 1 Etherium coin and was actively mining more, with a calculation power of about 2,100 million hashes per second. That’s roughly equivalent to the output of 85 computers each running a Radeon Rx 480 graphics card or 1,135 computers running a GeForce GTX 560M, based on figures provided here.”

A sane or at-all well-run society wouldn’t allow such a stupid waste of energy without a license. Why can any idiot with fake money waste so much power and heat to create more fake money in a currency that’s so volatile it yoyos up 2000% in a year and then plummets 50% in a month?

This is just a spectular waste of know-how, effort and energy that could be better expended elsewhere. Our stupid capitalist system encourages this kind of onanistic bullshit.

A thought experiment: what if this energy waste was a significant part of the energy budget of the Earth? Say 1%? What then? Would we regulate who gets to do this? The market-pricing doesn’t reflect the actual cost to the planet because the primitive measures imposed by our system simply don’t account for it. Pollution, environmental and ecological impact are all negligible side-effects absorbed for free by everyone while the thieves clean up.

From another article a few days later, In Iceland, bitcoin mining will soon use more energy than its residents by Cyrus Farivar (Ars Technica)

“We are spending tens or maybe hundreds of megawatts on producing something that has no tangible existence and no real use for humans outside the realm of financial speculation,“ Smári McCarthy, and Icelandic member of parliament told the AP. “That can’t be good.”

“Cryptocurrency mining requires almost no staff, very little in capital investments, and mostly leaves no taxes neither,” he wrote. “The value to Iceland / value-generated ratio is virtually zero. Closer to zero the higher the value of cryptocurrencies.”

This is absolute madness, a microcosm of the sickness affecting our society.

I noticed that when I emptied the trash in High Sierra, I didn’t see a corresponding rise in available space in the status bar of open Finder windows.

I opened “About This Mac” to the “Storage” tab. Here I could see that my System was using 70GB.

Huge system (not mine)

A search turned up the article macOS High Sierra – Huge System Storage (Apple Support), which explains that High Sierra introduced a feature called local backups for Time Machine (About Time Machine local snapshots (Apple)).

It does pretty much what you would expect it to do:

“Your Time Machine backup disk might not always be available, so Time Machine also stores some of its backups on your built-in startup drive and other local drives. These backups are called local snapshots. One snapshot is saved every hour. If you’re using macOS High Sierra, another snapshot is saved before installing any macOS update.”

And Apple swears that this feature won’t get in your way:

“Snapshots older than 24 hours are automatically deleted. And to make sure that you have storage space when you need it, snapshots are stored only on disks that have plenty of free space. When storage space gets low, additional snapshots are deleted, starting with the oldest. That’s why Finder and Get Info windows don’t include local snapshots in their calculations of the storage space available on a disk. (Emphasis added.)”

I’ve found that the local backups area reflected in the numbers shown in Finder windows. When I deleted local backups, one of them caused over 40GB to be recovered. But today, I found that the local backups had continued, but weren’t taking that much space. I think this feature just uses more apparent space when you’ve recently done a large system update (as MacOS has had of late). I’m not convinced that these backups are truly removed after 24 hours, though.

For even more space back, you can delete older iOS backups to get back GBs of space.

1. Open iTunes
2. Open Preferences
3. Select Devices (second-to-last icon)
4. Delete old backups

I had a good dozen gigabytes of backups from 4 years ago that I could delete.

Use Cases

This evaluation is for a machine that is excellent at the following tasks:

• Photo editing with Photos
• Playing HD movies
• Using Word processors like Pages and Word

The following tasks are not a priority:

• Playing high-powered 3D games
• Video editing

Requirements

• Must run MacOS [1]
• Must be relatively quiet (fan noise only in extreme usage scenarios)
• Long battery life (6+ hours)
• Lightweight/portable/easy to store in most bags
• Sturdy (doesn’t feel plastic-y or cheap)
• High-quality touchpad and keyboard
• High-resolution screen (i.e. “retina”)
• At least a solid I5 Kaby chip
• At least 8GB of RAM

Models

Since portability is a requirement, we’re going to take a look at two smaller models, leaving out the 15" MacBook Pro, which is more of a high-end, developer-style machine.

12-inch MacBook Pro

Pros:

• Runs MacOS
• No fan, so it’s dead silent
• 10-hour battery life
• Under 1kg and smaller than a sheet of Letter-size paper
• 12" Retina display with native 2304-by-1440 resolution
• Full-size keyboard and large touchpad with Force Touch
• 1.3GHz i5 (turbo to 3.2GHz) or 1.4GHz i7 (turbo to 3.6GHz) (both Kaby)
• Available in Space Gray, Gold, Rose Gold and Silver

Cons:

• Lo-res Webcam (480p)
• Only has one port USB-C port for everything—including charging. You have to dongle everything off of that.
• USB-C port doesn’t double as a Thunderbolt 3 port
• At 12", screen space is limited; you can increase resolution, but sacrifice readability. Only 81% of 13.3" MacBook Pro.
• Graphics card is weaker than MacBook Pro, but probably more than enough for a non-gamer

Macbook Pro

Pros:

• Runs MacOS
• Experience shows the fan comes on rarely, even under medium load
• 10-hour battery life
• Under 1.5kg and just a bit larger than a sheet of Letter-size paper
• 13.3" Retina display with native 2560-by-1600 resolution
• Full-size keyboard and giant touchpad with Force Touch
• 2.3GHz i5 (turbo to 3.6GHz) or 2.5GHz i7 (turbo to 4.0GHz) (both Kaby)
• Available in Space Gray and Silver
• Hi-res Webcam (720p)
• USB-C ports are also Thunderbolt ports
• Has an HDMI out that can drive a 5k display
• Speakers are pretty good for a laptop

Cons:

• Lo-res Webcam (480p)
• Has two port Thunderbolt/USB-C ports for everything—including charging. You have to dongle everything off of that.

General Comments

• The standard SSD size of 128GB is a joke if you have any movies, photos or songs. Anyone who’s been accumulating data for a while is going to need at least 256GB, if not 512GB. Apple’s default size is set with a cloud user in mind. Most home users will not need that much space, but 512GB is a good sweet spot for me (I have 82GB of pictures and 112GB of music).
• Not a heckuva lot of ports. The MacBook only has one. That said, you can get an adaptor or two for when you’re hooked up at home. They each cost about $50.– to $70.–, though.
• The screen size is small for developers, but just fine for everyone else, I think. MacOS accommodates full-screen apps really well, letting you quickly navigate between windows and desktops.
• The TouchBar looks interesting and having a fingerprint sensor would be nice, but I just don’t see this feature being used that much. I’ve seen it in stores and it seemed kind of interesting, so it’s up to the user.

Recommendations

You’re going to have to know for yourself, but here are the specs I would recommend for each model:

12" MacBook

• 1.4GHz Processor
• 512GB Storage
• 8GB RAM
• Multiport adaptor for USB devices/charging

13.3" MacBook Pro

• 2.3GHz Processor
• 512GB Storage
• 8GB RAM
• Multiport adaptor for USB devices/charging

Links

The following reviews were helpful:

• 13-inch MacBook Pro vs. 12-inch MacBook (2016)
• 12-inch MacBook vs 13-inch MacBook Pro with Touch Bar

Before [1] jumping on the Bitcoin bandwagon, you should acquaint yourself with how it works. I found the following video to be quite helpful.

Ever wonder how Bitcoin (and other cryptocurrencies) actually work? by 3Blue1Brown (YouTube)

Another article, How bitcoins became worth $10,000 covers some of the same territory as the video, but also provides more of the history of Bitcoin as well as details about some of the current factions and infighting among the various communities and e-currency proponents.

“Bitcoin’s status as the default medium of exchange in the blockchain world pushes its value upward in a similar way. Bitcoin is a popular medium of exchange for cryptocurrency-to-cryptocurrency transactions. […]

“This helps to explain why Bitcoin’s price started to soar in early 2017 at the same time as the larger blockchain ecosystem was booming. People needed bitcoins to participate in other cryptocurrency offerings, and Bitcoin became caught up in the general cryptocurrency euphoria. […]”

Finally, the article Bitcoin an even bigger waste of energy by John Quiggin (Crooked Timber) addresses the bubble-like nature and extraordinary wastefulness of making a currency based on “work” in a world already besieged by climate change.

“[…] most of the market value of a Bitcoin reflects the electricity wasted in the calculations needed to “mine” it, with the obvious disastrous implications for the global climate.”

This might not be so bad, if it was actually worth something. But it isn’t. People will make money off of it, but not because anything of value was actually produced in the real world. Bitcoins aren’t even really a currency—there are few places where you can use Bitcoins or any of the other e-currencies. That means that they are just another financial asset class that will make the riders of the bubble a lot of money and lose money for everyone else. It’s a pyramid scheme.

“Despite the huge increase in the market value of bitcoins, they seem further than ever from becoming an actual currency. Unsurprisingly, there’s no sign that governments are willing to accept bitcoins as legal tender. Nor is there any sign that they are displacing standard forms of money. On the contrary, bitcoins now seem to be seen as a financial asset, with no real suggestion that they will ever be a general medium of exchange.”

Quiggin, who is an economist, takes heart that, despite the waste of power, damage to the environment and strong likelihood that people are, once again, being duped, Bitcoin’s success would prove him wrong and prove EMH-supporters right.

“At any rate, the durability and magnitude of the Bitcoin phenomenon, running for nearly 10 years and with a putative value of nearly $US 100 billion, provides us with a very sharp test of the Efficient (financial) Markets Hypothesis. If Bitcoin eventually becomes a currency, the EMH and its supporters will be vindicated, and I (along with quite a few other economists) will have a lot of egg on my face. If the bubble bursts, the roles will be reversed.”

That Quiggin thinks that the “roles will be reversed” is far more hopeful than I’m willing to be: When the bubble bursts, nothing will change and we will all get ready for the next bubble, herded onward by the same cowpokes who drove us to financial ruin during the last one.

Some of the comments on that article offer some interesting insights.

This is a public-service announcement. There are several ways of blocking trackers and ads on desktop browsers. I’ve been using Ghostery for years, but am a little leery of the company behind it. For a while now, my browser of choice, Opera [1], has included ad-blocking natively. I’d also heard about a tool from the EFF named Privacy Badger that is supposed to do what Ghostery does.

I wanted to figure out which software I actually need to use in order to turn off tracking. To that end, the EFF provides a tool called Panopticlick that you can use to test how well your browser blocks trackers and respects the “Do-Not-Track” flag.

• Opera’s ad-blocker is fast and native, but blocks only “most” ads. You don’t see many ads when browsing with only that, but it doesn’t block any trackers. Opera has a “Do-Not-Track” flag that you can enable. Honest sites will honor it. Still mostly red in Panopticlick.
• EFF’s own Privacy Badger passes with flying colors. My browser still has a unique fingerprint, but I’m not sure if there’s a way of blocking that. All green in Panopticlick.
• Ghostery does not block as much as Privacy Badger. Mostly green, but one yellow/red check in Panopticlick.

Long story short: I uninstalled Ghostery and installed Privacy Badger instead.

Browser Setup

As an addendum, these are the measures I take for private browsing:

1. I use Opera because it has the features I use below.
2. Enable the “Send a ‘Do Not Track’ request with your browsing traffic” checkbox in preferences
3. Enable the “Block ads and surf the web up to three times faster” checkbox in preferences. I use the default settings and whitelists, but you may want to be more aggressive. Remember that a lot of web sites have advertising as their only income source. If they can’t make money, they might stop entertaining you.
4. Install the “Privacy Badger” extension with default settings.
5. Change your default search engine to “DuckDuckGo” to avoid Google tracking.
6. Turn on the VPN in Opera when you want to spoof your location/IP address. This is fast and easy and doesn’t interrupt or slow down normal browsing. It does involve proxying your traffic over another node (through the VPN tunnel), so your mileage may vary with video streams. I don’t use this very much, but it’s handy for when I need it.

I’d recently started using Windows Remote Desktop (RDP) to connect to my desktop at work from my laptop. The advantages are many: My nearly 4-year-old laptop no longer has to groan under the strain of running Visual Studio with 150 projects open. It was working very well, actually, but drained the battery considerably faster than I’d like. The battery lasts longer and the CPU no longer runs so hot, making it a viable laptop again. Hat tip to Björn for telling me to give it a go.

Unfortunately, it’s not all wine and roses: Windows 10 immediately added two new keyboards for me, keyboards that I neither need nor want nor have any use for. They are tangentially related to my settings—ENG (SG) and ENG (US)—but are not input methods. My input method is de-CH. Just because I’m running Windows 10 in English doesn’t mean that I need English input methods.

Not only that, but Windows 10 seemed to be randomly and very irritatingly switching between keyboards. The standard combination for switching keyboards is Win + Space but I was 100% certain that I wasn’t hitting that as often as Windows seemed to be switching keyboards.

Something was wrong.

But’s not a huge problem, right? Just remove the extra keyboards, right?

Right?

Would I be writing a blog post about it if it were that straightforward?

I would not.

As you can imagine, it’s not so easy to get rid of the extra keyboards because Windows 10 doesn’t admit that they exist. They were added when I first connected with RDP. The input-language menu is enabled, but there’s only one keyboard/input language in the settings.

What to do?

A colleague said he knew the solution and it is typically counterintuitive: add the keyboards to the list of input languages and then remove them again.

Genius, right?

This actually worked for one of the keyboards—ENG-US. It was no longer in the list of input languages and the entry on the input-language menu was gone, as well.

However, there is no entry for ENG-SG that I can add and remove.

Not only that but, as soon as I connected with RDP that night, Windows 10 replaced the “missing” keyboard and I was back to square one.

And I’m not going to start editing the registry to fix input languages. That feels like utter capitulation.

So I can’t get rid of the keyboards.

But windows 10 keeps changing my keyboard and I’m not doing it. Spoiler alert: Windows 10 has another key mapped to “change input language*: Left Alt + ⇧.

This one is a gigantic pain in the ass. I hit this key combination all the time.

So, I can’t remove the keyboards and I can’t stop using that key combination (it’s intrinsic to text navigation), so I have to figure out how to get rid of that key combination.

Here’s how to do it. There are rather more steps than one might expect.

1. Click the language menu on the task bar
2. Choose “language preferences”
3. Choose “additional date, time and regional settings”
4. Choose “change input methods”
5. Choose “advanced settings”
6. Choose “change language bar hot keys”
7. Choose “change key sequence”
8. Change key for “Switch input language” to “Not assigned”

See the screenshots below for context.

In the several days since I’ve done this, I have no longer inadvertently switched input languages. I still have three keyboards, but I don’t have to care anymore.

1. Input language

2. Region and language settings

3. Clock, language and region

4. Language settings

5. Advanced settings

6. Text services and input languages

7. Change key sequence

Every once in a while, I feel bad for not using the only remaining truly open-source browser: Firefox. [1]

Opera is my go-to browser, but that’s been bought up by the Chinese, right? So far, things haven’t changed but who knows what’s going on under the hood? Chrome is a decent testing browser, but offers me nothing better than Opera and its memory profile is, for whatever reason, consistently worse. I’m a very happy Opera user.

Still, Firefox is, in version 57, supposedly much faster and uses less memory, finally putting it on-par with the other browsers. Cool. Let’s check it out.

Y U no speak English?

I start Firefox and it thinks I want to speak German. It’s a mistake more than a few applications make. My location is Switzerland, but the language of my operating system is English. Why does Firefox choose German instead of English? It’s not alone, but it’s a boneheaded mistake—if you’re going to make a guess, then guess the language I’ve chosen for my whole operating system.

No problem, though. I’ll just go find the setting where I can change the UI language. [2]

Y U no have UI-language setting?

There’s no way of searching preferences.

I guess I’ll just click through every page of the settings, hoping to find the language settings.

Damn, I can’t even find the area where I could change the list of preferred languages for web pages.

Y U no Upgrade?

Which version of Firefox am I even using?

Wait, how the heck do I figure out which version of Firefox this is? Let me type about:. Oh, that doesn’t work anymore. Back to the kind-of confusing menu on the right-hand side. None of those little pictures look right … maybe if I click “help”? Ah, there it is.

Apparently, I have an update pending. I’m surprised that wasn’t more prominently displayed.

Restart Firefox to apply updates.

I have a new start page and new settings! Now I can find the list of preferred languages for web pages, but still nothing about the UI language.

Y U no Helping?

I guess I’ll have to give on Firefox helping me and will ask DuckDuckGo.

I find the article Firefox in your language? Here’s the tutorial. Apparently, it’s possible without re-installation, for which I suppose I should be grateful…but there’s a tutorial. Sounds ominous.

I have to install an add-on. Whatever. I’ve already sunk enough time; let’s get this done.

I jump to the link provided for language packs and the one for English is right at the top.

Flips Table

How is that even a thing? How is it possible for Firefox to release a version on their stable branch without having the English language pack available? I’m running the latest stable version.

It was already a long-enough slog to get here in the first place, but now I find out that it’s not even possible to change the language of my browser’s UI without re-installing? Or that I get to “Download anyway” and take my chances?

Once again, I admit defeat and stop using Firefox. Who cares how fast it is [3] when the user experience is still so balky? It continues to feel like an unpolished product.

The Web Annotation Working Group has published [t]hree recommendations to enable annotations on the web.

What does that mean for you, as a consumer of the Internet?

This standard will bring commenting and conversation and annotation into the 21st century. It will possibly bring order to the myriad systems and accounts and formats currently in place that allow for interaction and discussion.

The diagram Web Annotation Architecture (W3C) (SVG) is interactive and steps you through the whole concept. It’s quite easy to follow and worth your while if you participate in conversations on-line.

Annotation Guide (SVG)

What this annotation standard doesn’t do is address the problem that most people are without facts, have little to no reasoning capacity, have big, dumb opinions, and even bigger mouths. No-one can fix that.

Separating annotations from content

When you read an article online, you currently see comments only from that site’s annotation system. While the site owner can moderate the comments, you cannot. If there are 4000 garbage-filled comments, you have to plow through them if you want to be involved in the discussion. There are various mechanisms for voting and rating to try to get useful content more attention, but it’s still often a burden.

The standard aims to put more power into the hands of the consumer by allowing users to display annotations that come from other sources. The idea is not to have a centralized provider (as it is now) but to allow a user to choose which annotation providers they want to enable on their browsers or on certain sites.

Under Your Control

Imagine if you had an annotation provider of just you and a circle of trusted confederates. When you went to a site, instead of seeing thousands of annotations from the unmoderated horde, you could elect to see annotations made only by members you trust.

Not only that but the disparate and ad-hoc annotation systems, each with different authentication and authorization requirements as well as formats could now be consolidated by the user (at least to some degree).

You could maybe even pay a few bucks a month to house your own annotations and re-publish them to other systems, like Facebook or Twitter. If those sites also provided annotations support, then you’d have your own copy of your annotations, but other people would see your annotations if they used the more popular providers from Facebook or Twitter. You’d have the best of both worlds.

If you use a Kindle reader or Instapaper (or both, like me), then you already have two annotation systems available. The Kindle stores to a text file, while Instapaper stores to a proprietary store online. Now imagine if your annotations for these systems were standardized and stored separately from both sites. You would make notes on the Kindle and they would be synced to the same place as notes made while visiting the page online or while reading in Instapaper. This sounds very good to me, at any rate, or probably to anyone with an academic bent.

Available Separately

The nice thing about this standard is that there is no need for buy-in on the part of the content providers.

Although a content provider can provide integration, a user is free to enable one or more external annotation systems. The user is free to do that today but the new standard provides a way of interaction between them that would allow this whole market to gain momentum.

If this idea were to spread, then annotations would become a separate cottage industry, with its own market and business model. Online news providers would no longer be obligated to waste time moderating and upgrading and protecting their own annotation/commenting systems. They could focus their time, effort and resources on journalism—or whatever passes for journalism at the New York Times.

With this separation, a user could pick content separately from the quality of the annotations. There would no loner be a notion of a siite’s content being good, but the community bad. If you don’t like the community, switch annotation providers, seek one out with better moderation. Imagine if, instead of having YouTube commenters annotation videos, you could enable (for example) the annotations from /r/truefilm or /r/movies on Reddit instead. If Reddit provided annotations support to this standard, then this could happen automatically whenever you browsed to a YouTube video for which there was a Reddit thread on one of those sub-reddits.

A common UI is easier to use and manage.

Another issue to address is that the annotations show up only at the end of the article, with no way of referencing a location in the content. Some annotation systems don’t even have threading, so comments are just stacked with no relation to one another. A common system would provide the same powerful UI regardless of what the content provider was willing to include in the web-site software.

There is the issue of scale: what if the original text is overwhelmed by annotations? That shouldn’t be a problem as you can turn annotations off at any time. If you have multiple sources, you can toggle them individually. Also, most annotations are replies to other annotations, so the number of root-level annotations is much smaller than the total and shouldn’t ever crowd out the content.

The threaded nature of annotations combined with control over annotation sources should ameliorate these issues. It’s certainly more power than you have today. [1]

Technical details

The working group has now officially been replaced by the Open Annotation Community Group. With this announcement, the data model, protocol and vocabulary have been approved by all stakeholders.

• Data model: “JSON format for ease of creation and consumption of annotations based on the conceptual model that accommodates these use cases”
• Protocol: “describes the transport mechanisms for creating and managing annotations in a method that is consistent with the Web Architecture and REST best practices”
• Vocabulary: “the set of RDF classes, predicates and named entities that are used by the Web Annotation Data Model”

That this concept has been standardized is a very big deal if we can actually make it happen. The standard would allow not only vendors to develop their own systems but browsers to provide native and fast implementations that use a common UI language in the browser itself.

A working implementation

If you click through to the article Annotation is now a web standard (Hypothesis), you can see a non-standard annotation system in action. It looks quite nice. The article describes the advantages of consolidation as well, cited below.

“While many applications, from PDF readers and Google Docs to the Kindle, support some kind of annotation functionality, what the W3C formalized yesterday is fundamentally different. The W3C architecture provides for a model where annotations live separately from documents and are reunited and reanchored in real-time whenever the relevant document is present. The benefit of this is that annotations now come under the control and election of the user, rather than at the sole discretion of the publisher.”

The existing plugin in the linked article already exists and looks very nice and subdued. The diagram below shows the initial view.

Embedding Annotations

Annotated text is highlighted, but you can toggle its display (with the “eye” icon on the right) for distraction-free reading. Controls on the side allow you to toggle the annotations panel and also show you a preview of the number of annotations available so far.

Scrolling down shows more annotations, below.

Annotations Visualization

As you scroll, you can see how many annotations are available above and below the current point. These indicators help you locate areas that are being heavily discussed or annotated. Of course, using these to navigate without reading the whole article leaves you susceptible to manipulation, but it can still be a good feature if used wisely. [2]

If you select an annotation in the text or the scrollbar or by clicking the panel-toggle, the panel opens on the right.

Annotations Panel

Having annotations to the right (or left) is a standard practice in much professional review or annotation software or printing (see Microsoft Word) and makes much more sense than below an article or paper.

Annotations are generally made in reference to a specific location in the text and should be in-line, so reader have context. This implementation does that and shows how browser vendors could build on this standard to improve the commenting experience considerably. The version shown above is clean and simple and powerful and enhances the article, as annotations should.

As content grows, the system grows with it, as shown below.

Threaded Replies

Threaded replies are not shown by default, but can be shown with a click. Each annotation has its own unique identifier and can be shared and perma-linked. A system can use color hints—or maybe icons or gravatars or subtle name-tags—to allow the reader to differentiate annotations by user.

I think this is a great advancement and look forward to its adoption and expansion.