Ran Prieur

February 27.

2026-02-27T15:30:06Z

February 27. Some thoughts on AI fakery, starting with the popular prediction that AI images will get so good that nobody will be able to tell real from fake. This is not going to happen. Intelligence agencies, the legal system, and journalists are well aware of this problem, and they will always have tools to stay one step ahead of fake images, if they care to use them. Meanwhile, people who want to be fooled can manage to be fooled by much more primitive tech.

Also, only one category of AI images presents this problem: images designed to look like photographs. People are obsessed with images designed to look like photographs because they're obsessed with the tension between fake and real. I'm obsessed with aesthetics. For my videos, I don't use photorealistic images because at the free level they look terrible. This makes me think that the future of video games will be obviously unreal images, because they're a much better value.

When you think about it, before the invention of photography, it was impossible to conceive of images as being true the way we do now. Even super-realistic Renaissance paintings were understood as the interpretation of the artist.

Meanwhile, all this time, there has been a free and potent technology for fakery. It's called words. All a set of words has to do, to be accepted as true, is to say "this is true." This is why AI images are so wild, while AI text is so constrained. Words are very dangerous, and the danger is growing. What I'm wondering is not "how long until the machines take over?" but "how long until the machines make humans too crazy to maintain the machines?"

February 24.

2026-02-24T12:00:25Z

February 24. Great article from Aeon, Books and Screens. The idea is, screens are making us stupid not because they're screens, but because "the dominant platforms have been deliberately engineered to fragment attention in service of advertising revenue."

They sit with laptops open, seven tabs competing for attention, notifications sliding in from three different apps, phones vibrating every few minutes. They're trying to read serious material while fighting a losing battle against behavioural psychology weaponised at scale. They believe their inability to focus is a personal failure rather than a design problem. They don't realise they're trying to think in a space optimised to prevent thinking.

During "refinement", the model gravitates toward the center of the Gaussian distribution, discarding "tail" data - the rare, precise, and complex tokens - to maximize statistical probability. Developers have exacerbated this through aggressive "safety" and "helpfulness" tuning, which deliberately penalizes unconventional linguistic friction. It is a silent, unauthorized amputation of intent, where the pursuit of low-perplexity output results in the total destruction of unique signal.

When an author uses AI for "polishing" a draft, they are not seeing improvement; they are witnessing semantic ablation. The AI identifies high-entropy clusters - the precise points where unique insights and "blood" reside - and systematically replaces them with the most probable, generic token sequences.

This is another example of a whole technology being blamed for a bad choice in using the technology, and this kind of ablation is not new and does not require high tech. It's exactly what happens in movie test screenings, or in almost any communication where the priority is to maximize the size of the audience.

Also, this is why I like the DeepAI image generator. It has 122 styles that make this choice in different ways. Some of them are slop-optimized, and some of them are instructed to do weird shit. Surely it wouldn't be hard to make a chatbot site that had the same variety of options. I think the trouble is that we're more afraid of text.

Finally, something lighter: I Taught My Dog to Vibe Code Games

February 20.

2026-02-20T20:20:08Z

February 20. Taking a break from techno doom for some woo-woo links. New theory proposes that consciousness is the universe's foundation, not atoms. Yeah, this is not a new theory, and this version doesn't go far enough. They're thinking, first there's the universal field of consciousness, then there's the big bang, galaxies, stars, the earth, humans, the brain, and finally you. I'm thinking, first there's the universal consciousness, then there's you, a fragment or aspect of the universal. Then there's your environment, which is not there until you pull it out of the infinite range of possibilities into this specific thing: humans, the earth, the sky. Then, out of the still wide range of possibilities for the sky, we pull out stars, galaxies, and finally the big bang. Just one step back and we could have an eternal universe in which quasars are spat out of galaxies like seeds.

This is the best NDE thread I've seen on Ask Reddit, People who have died briefly, what did you hear, see, and feel?

Getting weirder, a sub-thread from a thread about unbelievable experiences, with several reports of flipping reality instead of dying

And an unpopular subreddit, TheTruthIsHere, "a database for non-fiction encounters with the unknown from a personal source."

February 16.

2026-02-16T16:40:04Z

February 16. AI links, starting with a scary Hacker News thread, An AI agent published a hit piece on me. From the top comment: "The AI companies have now unleashed stochastic chaos on the entire open source ecosystem. They are 'just releasing models', and individuals are playing out all possible use cases, good and bad, at once."

I've written before that one thing AI might do, that I would be excited about, is video game worlds that are as creative as human-made worlds, but with no edges. That's looking unlikely: AI can't make good video game worlds yet, and it might never be able to.

Last year I mentioned a severe AI personality called Absolute Mode. That's Keith's blog post on it, and this is his latest post about wrangling with ChatGPT to actually do it. What I think I understand, is that it is possible, but somewhat difficult, to give an AI a custom personality. If so, this is going to be huge. I mean the whole bubble might pop any day, but if not, someone is going to make a lot of money from slicing and packaging chatbot personalities for mass consumption. Give me a sassy bitch. Give me a Klingon. Give me a golden retriever. Or maybe we won't go down that road because too many people want to talk to a cult leader.

Isn't it funny how talking computers turned out? In old-time sci-fi, they're rational, robotic, precise, and never wrong. Instead, they're like goofy sidekicks, offering encouragement and ideas, but clumsy and unreliable. We thought we were getting the Professor and we got Gilligan.

February 13.

2026-02-13T13:10:10Z

February 13. Today, 2025 films. There were two sprawling political thrillers, both technically well made by respected writer-directors. Paul Thomas Anderson's One Battle After Another grossed over 200 million, got rave reviews, and will probably win Best Picture. It is Hollywood bullshit. Ari Aster's Eddington grossed under 14 million, got middling reviews, and no Oscar nominations. It was the best film of the year.

Sean Penn will probably win best supporting actor for playing the villainous Colonel Lockjaw, and it really is great acting. But that character, like every character in One Battle After Another, is a cartoon with zero moral complexity. Meanwhile, Joaquin Phoenix in Eddington, and Jesse Plemons in Bugonia, got snubbed for awards by playing complex and fully human red tribe characters.

I'm avoiding the words right and left, but Eddington gave me a sense of how a person might be tagged as "right" for finding what passes for the "left" unbearable: cringey teens and politicians who are tools of big money. Also, Eddington is tagged as a satire, but what I see is authenticity. Reality itself is absurd, especially political reality, and any work of fiction that shows it accurately will seem satirical, while any work of fiction that plays it completely seriously will be propaganda.

Some other 2025 films. Most overrated: Frankenstein, which is flat out just a superhero movie, with great set design, total bullshit dialogue, and they even gave the monster super powers. Most underrated: Rabbit Trap, 4.8 on IMDB but it's a solid 7, a weird horror/art film about fairyland. Most enjoyable: Companion, a comedy thriller about a weaponized sexbot. Least enjoyable: Marty Supreme, in which Timothee Chalamet plays the most unlikeable protagonist of all time -- even Jake Gyllenhaal in Nightcrawler was admirable for being cool-headed. Best horror film and second best film of the year: Bring Her Back.

February 11.

2026-02-11T23:50:37Z

February 11. Not in the mood to post this week. Three Reddit links: What's a book that you feel encapsulates your soul and why?

Couples of Reddit, what's the dumbest 'house rule' you and your partner made as a joke but now both of you are low-key aggressively serious about enforcing?

What weird skill did you accidentally become good at because of a bad job, hobby, or situation?

February 7.

2026-02-07T19:10:56Z

February 7. One more link on the below subject, a PsyPost article on cognitive debt:

Electroencephalography (EEG) monitoring revealed that participants in the ChatGPT group showed substantially lower brain activation in networks typically engaged during cognitive tasks. The brain was simply doing less work. More alerting was the finding that this "weaker neural connectivity" persisted even when these participants switched to writing essays without AI.

February 6.

2026-02-06T18:00:29Z

February 6. Some links about technology and human cognition. Outsourcing thinking is a Hacker News thread with a linked blog post, a careful analysis of the many ways that our own thinking is compromised, if we let LLMs do our thinking for us.

Wirth's Revenge is another Hacker News thread with a linked blog post, this one about programming, and how exactly AI is good and bad at it. From the post: "LLMs don't do reliable, they don't do repeatable.... You don't ask the LLM to perform a repetitive and precise task, you ask it to build a script that performs that task. Except in rare cases, this script does not itself use LLMs."

An archive of a piece from the Atlantic, The Film Students Who Can No Longer Sit Through Films

A deleted Reddit comment, with a long thread below it, about the vanishing middle among grade school students, how there are a few who are really smart because their parents got them reading at a young age, and then a whole bunch who are functionally illiterate because their parents parked them in front of screens.

Also from Reddit, an explanation of P = NP. Basically, even though we can't prove it, we know that P does not equal NP, because coming up with a solution is in a different mind space than figuring out whether a solution is correct, and the more complex the problem, the greater the difference. There's a long comment thread including some stuff about teaching, and how someone who is naturally good at something is the worst teacher, because they can't get in the mind space of someone who finds it difficult.

For me, the difference between finding a solution, and recognizing a solution, bears directly on AI images. If I want a good image on a given theme, it's much easier to have an AI generate a bunch of images and pick the best one (the below image was picked out of about 40) than it is to actually draw an image that's that good. So I save a lot of work, but by taking that road, I am failing to gain an understanding of how to draw. Life is short, and I'd rather not know how to draw than not know how to think.

February 3.

2026-02-03T15:30:22Z

February 3. I don't want to write about Trump and just say the obvious stuff. So here's something you don't hear every day: Trump is a scientist, and a good one. For his entire career, not just the presidency, he has been methodically doing experiments on human institutions and the human psyche, to see how they stand up to raw power. The most honest thing he ever said was during a debate with Hillary Clinton, where she accused him of tax evasion and he said something like "It was your job to stop me." That was Trump announcing to the world that he is beyond good and evil. He has the ethos of a fire or flood. ICE is not arresting the most dangerous immigrants, but the most compliant, for the same reason a flood fills in the lowest places first.

If it's our job to stop Trump, how are we doing? Imagine you're a teacher giving out letter grades. Around 30 percent of Americans are currently riding an F. With the long memory of the internet, it's going to be hard for anyone openly supporting the Minneapolis killings to walk it back. (And how weird is it that their names are Good and Pretti?) Europe gets a C for eventually standing up for Greenland. Congress and the Supreme Court might yet squeak out a D. If anyone gets an A it's the people quietly fighting in the courts.

The American media get a flat F. The experiment is how much naked power does it take for them to report it as naked power, and the answer is we don't know yet. Their expectation of normalcy has been Trump's number one ally, the fog of war of the supposedly unthinkable. They swallowed Trump's tale that Greenland is a buffer against Russia, when his actions have made it clear that Europe is his adversary and Russia is his uneasy ally. I used to think propaganda meant lying. Now I know that propaganda is saying stuff that's technically true while never saying the most obvious and important stuff.

Trump's obvious electoral strategy is to stir up enough trouble in the cities to cancel midterm elections. If he tries, the states will turn the tables: We're still having elections, try to stop us. MAGA will have to defend federal power over states rights, further pushing the experiment of how much cognitive dissonance the human brain can tolerate. We don't know yet.

Domestically, Trump doesn't have a lot of cards left to play, which is why he's focusing on Iran, and I appreciate the honesty of his foreign policy. No more iron fist in velvet glove, it's all iron fist, fuck you. If somebody goes nuclear, all bets are off. If not, I expect this to be over soon. Historians will surely say that America caught the same disease as Nazi Germany, but that we resisted it better. I know there are full-on concentration camps right now, but I am confident that Trump's eventual death toll will be less than one percent of Hitler's death toll, unless you count the global poor killed by canceling USAID.

I'm confident because I live in the city. If you don't live in an American city, you might have the idea, from the news, that we're all delicate intelligentsia and gutter trash. What I see are tens of thousands of competent, determined, and cheerful people. Even the homeless are tougher than ICE, give them guns and you'll find out. This feels to me like a very slow mass shooting. The shooter is inside the building, he has the upper hand at the moment, and he's not finished. But he can't win, that's not how this works.