Guest post by Alexander "Sasha" Putilin

[This is a guest post by 2024 ACX grantee Sasha Putilin. I encourage any ACX grantees who are interested to write about their projects. - SA]

The results of my ACX Grants 2024 project are in.

The project attempted to replicate the 2023 study "Learning at your brain's rhythm: individualized entrainment boosts learning for perceptual decisions". It claimed that if you read a person's brain waves, figured out an individual peak alpha frequency, and flashed a bright white light at that frequency, then they learned a certain perceptual task faster.

Why bother? The result hinted that learning may depend in part on how well the brain keeps its rhythms coordinated. In other words, perceptual learning may rely on an internal brain metronome. If flickering light could act as an external metronome, it might help the brain maintain the right rhythm and learn faster.

The study offered an invitation to develop new frontiers of neuroscience and biohacking. If the effect generalised to other types of learning, you could build a learning helmet: put it on your head, let it read your brainwaves, flicker light tailored to your individual brain — and you learn a new skill quicker.

And no, it didn't replicate. Most likely it can't replicate, because the effect is probably not real. The original study obscured the data with summary statistics. Running a $32,000 replication was excessive. We could've caught the issue with this study if we simply looked at the original data carefully.

*record scratch* *freeze frame* Yep, that's me. You're probably wondering how I got here. Here's the story.

https://www.astralcodexten.com/p/never-cross-a-river-four-feet-deep

I recently had a minor spat over someone misinterpreting my AI beliefs (see section marked "Update" at the bottom here), so I thought I would list them in one place, so I can refer people when they ask.

Define AGI as AI intelligent enough to do 90% of knowledge work jobs. I think there's a 25% chance of AGI by 20272, a 50% chance by 2034, and a 75% chance by 2045.

Basic argument: In a certain sense, AI is already "smart" enough for this (eg it can answer quantum physics problems, which require higher IQ than most knowledge work). Its remaining limitations are that it's confused, unagentic, lacks situational awareness, and tends to hallucinate. The METR time horizon graph, and several other related benchmarks/experiments/intuition pumps, suggest it's improving on time horizons at an (exponential) rate that lets it cross human-level performance sometime around the early end of the schedule above, and subjectively it feels like harder-to-measure constructs like situational awareness are improving about as fast.

Arguments for earlier: recursive self-improvement causes a speedup compared to the trend. This is one of the biggest blank spots in my model: I don't know how fast RSI will progress, and I don't think anyone else does either. There's some function mapping a combination of AI talent and compute to progress, and we don't know how it behaves in the domain when there's far more talent than compute available. It could fizzle out completely for lack of compute, or it could go vertical. The AI Futures Project has done some of the best work trying to model this, but even they have low confidence.

The philosophers of the Frankfurt School practiced a technique called negative dialectics, where concepts are defined as much by what you can't say about them as what you can. Appropriately, the Frankfurt School has ended up defined by what you can't say about them.

You can't say that they invented a new form of left-wing thought called Cultural Marxism. This would be (according to Wikipedia) the Cultural Marxism conspiracy theory, a "far right anti-Semitic conspiracy theory that misinterprets Western Marxism, especially the Frankfurt School, as being responsible for modern progressive movements, identity politics, and political correctness". You're not supposed to dub them a transitional stage between Communism and postmodernism. You're not allowed to speculate that a lot of the academic humanities, as they're practiced today, descend from the Frankfurt School's brand of critical theory. You're not supposed to think of them as the point where the muscular pro-technology leftism of the early 1900s shattered into the pessimistic degrowth leftism of the present.

Art is long, life is short. Most of us only manage to not do a few things in our limited span on Earth. But the Frankfurt School managed to not invent so many movements - to not be involved in so many of the crucial ideological shifts of the past century - that they caught my attention. Who were these people? What other aspects of our culture might we be unable to say they were involved in? For answers, I turned to the classic history of the group, Martin Jay's The Dialectical Imagination.

The basics are simple enough: the School was founded in Frankfurt in 1923. It attracted great philosophers like Max Horkheimer, Theodor Adorno, and Herbert Marcuse. When the Nazis took power in the early 1930s, the mostly-Jewish Frankfurters fled to America, where friendly locals helped them continue their work in affiliation with Columbia University. Mid-century Americans were suckers for sophisticated European intellectuals, and when the rise of fascism and World War II started dominating headlines, the German-Jewish Frankfurters were natural experts to help Americans process the situation. By the end of the war, they were firmly established as thought leaders. Some - including Horkheimer and Adorno - returned to Germany to rebuild its intellectual culture from the ruins; others stayed in America and remained relevant through the 60s and 70s.

But figuring out what the Frankfurters believed is more complicated. Forget about the thin line between universally-acknowledged fact and fascist conspiracy theory. The School itself was famously coy, worrying that if they explained themselves too clearly, people would caricature their beliefs and integrate them into the existing capitalist system. Even when they did speak "clearly", it was in the sort of German philosophical register where "the negation of the negation" is a totally normal thing to say.

Having only read a single book on them, I will no doubt fall into all the failure modes that they and their successors warned us against. But here are the analogies, intuition pumps, and parables that I found helpful.

https://www.astralcodexten.com/p/book-review-the-dialectical-imagination

I'm not saying AI is superintelligent or can decide better than you can. I'm saying that if you - like me - spend an hour or so doing research before voting on local seats, AI can aid that research very effectively. And if you don't do that research - because you weren't willing to waste an hour on it before - AI makes it so much faster that you might want to start.

I gave Claude a prompt something like (edited for coherence):

I'll be voting in the June 2026 California primary. I'm a centrist liberal abundance YIMBY whose favorite political writers are Kelsey Piper, Matt Yglesias, and Ezra Klein. I'm wary of government overreach, but I'm not a doctrinaire libertarian and want to help people when we can figure ways to do it that work. I'm going to ask you about each race on my ballot, and I'd like for you to list the various candidates' bios, policies, endorsements, your read on the most important differences between them, and your advice for me as I try to make my choice.

…and got back answers like the following:

https://www.astralcodexten.com/p/use-ai-this-election

Sorry, I give up.

In past elections, I've covered every single candidate for governor of California, from the incumbents all the way down to the cranks. In 2022 there were twenty-six of them, and I covered them all.

But sorry, I give up. This year there are sixty. It's too many. I can't disambiguate them all into unique individuals with their own personalities, hopes, and dreams. So as consolation for the list I'm not giving you, here are the basic types, and a few examples of each.

One of these people will definitely win, but what else is there to say about them? They're all the same. They've all paid the danegeld to some set of unions and interest groups, then put up some kind of incredibly generic platform about how they're compassionate but also a fighter. I can't bring myself to name any of them or discuss them further.

More or less the same as the top-tier Democrats, minus the chance of winning. They all have cowboy hats and flag pins. They all pose on horseback at their ranch. They all promise to Take Back California for the forces of America and its god, Donald J. Trump. None of these people are actually interesting, but honorable mention to Sheriff Chad Bianco, whose name is Italian for "white Chad". He might be the perfect Republican candidate:

https://www.astralcodexten.com/p/the-types-of-candidate-you-find-in

"All exponentials eventually become sigmoids" is an annoying AI talking point. If someone presents a graph like this…

….and points out that it seems like AI capabilities could soon reach the level marked "High", then the height of intelligent debate is to point out that actually, the trend could go like this:

…and then it would never reach the level marked "High"!

In slogan form, this is "all exponentials eventually become sigmoids" (a sigmoid is the s-shape of the second graph, which starts exponential but gradually flattens out). It's technically true. No process can keep growing forever; eventually it hits physical or practical limits. For example, total cases during an epidemic is classically sigmoid:

In conclusion, the only good theory of taste is Nostalgebraist's.

He wrote a post called Hydrogen Jukeboxes, analyzing the literary output of an AI called R1. This AI tried hard to write good fiction, which was part of the problem. It crammed its stories with what Nostalgebraist called (stealing a term from Ginsberg) the "eyeball kick" - a flashy stylistic move that immediately catches the reader's attention and "wows" them. Here are examples - some from R1, others from an experimental OpenAI model trained specifically for fiction-writing:

• "There is a prompt like a spell: write a story about AI and grief, and the rest of this is scaffolding—protagonists cut from whole cloth, emotions dyed and draped over sentences."
• "When the jar of Sam's laughter shattered, Eli found the sound pooled on the floorboards like liquid amber, thick and slow. It had been their best summer, that laughter—ripe with fireflies and porch wine—now seeping into the cracks, fermenting."
• "And so I built a Mila and a Kai and a field of marigolds that never existed. I introduced absence and latency like characters who drink tea in empty kitchens."
• "The morning her shadow began unspooling from her feet, Clara found it coiled beneath the kitchen table like a serpent made of smoke."

Nostalgebraist and another writer, Coagulopath, catalogue some of the most common AI eyeball kicks, each occurring across multiple LLM models:

• "An overwhelming reliance on cliche. Everything is a shadow, an echo, a whisper, a void, a heartbeat, a pulse, a river, a flower—you see it spinning its Rolodex of 20-30 generic images and selecting one at random."
• "Conjunctions combining one thing that is abstract and/or incorporeal with another thing that is concrete and/or sensory."
• "Repetitive writing. Once you've seen about ten R1 samples you can recognize its style on sight. The way it italicises the last word of a sentence. Its endless "not thing x, but thing y" parallelisms…the way how, if you don't like a story, it's almost pointless reprompting it: you just get the same stuff again, smeared around your plate a bit."

https://www.astralcodexten.com/p/nostalgebraists-hydrogen-jukeboxes