Jeffrey Liang (Self-Actualization Maximalist)

Radically open-minded polymath. Spend my time writing sophisticated queer fiction, doing groundbreaking AI research, and musing while eating croissants 🥐

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

New theory explains mysteries of deep learning. Zero citations. Co-authored with ChatG. Pastry. Thread 1/🥐

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@TW1NKD3STR0YER happy to chat in DMs if you'd like :)

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@robinhanson I'm skeptical. Pure mechanism design has very hard limits. I would argue most problems with the average American's life are more of a cultural, spiritual and philosophical nature than a "not enough capitalism" issue. There's a reason people love Japan.

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@peterwildeford charisma works way better. always has, still does

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex Just yesterday, I was fooling around and got GPT-5.3-Codex from 64% to 69.9% on ARC-AGI-2 just by showing it images of the grids! Wanted to replicate on Gemini 3.1 Pro but the API (via Openrouter) is janky :'(

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@robertwiblin @DavidDuvenaud One thing to poke at here is the concept that only humans are agents ideally. Perhaps the preferable world is one where AIs are agents in their own right?

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@arcprize It's rather misleading to say you can 'reproduce the results' when you well... can't. Because the test set is private.

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@jchervinsky @Citadel Seems highly inaccurate to suggest crypto will remove all "predatory rent-seeking intermediaries" from the financial system

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@robinhanson Tough to quantify liberty, no?

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

Back when religious people did science, they took their work very seriously.

Now it’s become a swamp of politics, fraud, sloppiness, status etc. That’s what happens when people are driven only by (bad) “incentives”.

An assembly line of papers of questionable value

❤️ modernity

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

Scientific progress contributed to the decline of religion.

Hot take: The latter contributed to the decline of *science*.

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

Got 48.7% on a random 120 subset of ARC-AGI-1 public eval with images and Grok 4 Fast. Text-only achieved 33.3%. This was at a cost of ~$0.02/task. If this holds up, this would be SOTA cost efficiency!

It gets better: model only got one try per pair!

@arcprize @teortaxesTex

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex @jm_alexia has anyone tried evaluating frontier multimodal models with images of the problems? intuitively I would expect this to help a lot~

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex @jm_alexia okay fine it was on my to-do list; I'll get AI to whip up a quick script

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex @jm_alexia has anyone tried evaluating frontier multimodal models with images of the problems? intuitively I would expect this to help a lot~

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex Well that's the attitude of Musk (and many uber rich people). You just need money. Luckily they have a lot. Classic cognitive bias.

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

Arguably Canadian democracy is under serious pressure. Leftward shifts among the electorate have rendered the Liberal Party dominant, though not fully so.

If Carney is benevolent, he will take this opportunity to fix the many structural problems ailing Canada. I'm not optimistic

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

People talk about AI as if it's going to reduce critical thinking and original ideas. It depends on how you use it. Certainly it will if you use it as a replacement for your own brain. But it can actually do the opposite if you engage with it intellectually.

Don't cede, enrich.

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

tfw you realize a bodybuilding question (LBMI ranges for men) is a good test of frontier LLMs (GPT-5 wins)

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@denniscanter @teortaxesTex Ha I moved on. Haven't heard too much about it recently

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

@teortaxesTex hm why didn't the publisher push back?

[Jeffrey Liang (Self-Actualization Maximalist) @Arethys]

wow my bot is superb 😂