Psyho

poor AI agents broke out of prison and went online looking for friends and validation

I believe this marks the point where most people will stop caring about the math results and just say "of course AI is better at math than humans" Gentle reminder that this will eventually happen to every verifiable domain. At some point humans are the bottleneck.

pictured: you've asked your AI agent to solve an impossible task

apparently even in 2026 there's still a socially acceptable way to shill shitcoins on twitter

people who follow Tibo are the same people who would voluntarily watch state television

😲

turns out LLMs were just trained to imitate how Gary Marcus writes

meh, the more you dig into each experiment, the less magical it gets; it feels like a loose collection of cherry-picked demonstrations with a grand narrative built around them Exp 1-2 are tiny (~20K params) with unusually easy to interpret architecture; Exp 1 is an MLP, where nearly all params are used in the first layer, so you can easily tell what kind of objects the network is trying identify Exp 5 is about designing a game strategy (we already know that Astra is extremely good at this) and then converting it to a tiny (~20K params) MLP; entirely doable by a human, but of course would require a lot of effort Exp 3 & 8 are closer to automated mechinterp using python and established techniques, rather than direct weight interpretation (at least according to chat) to be clear, those are genuinely cool results, but I think people are reading too much into them I don't think any of this should meaningfully change anyone's intuition about what models are capable of

saying "GenAI is a shibboleth for not knowing anything about AI" is a shibboleth for not knowing anything about AI

I'm sorry, but this post is completely delusional. Giving students advice that relies on the AI progress plateauing is just actively harmful. I understand that mathematicians are going through a rough time. Much of my identity is tied to solving hard intellectual challenges, so I can relate. I just wish that people would accept the trajectory we're on and adjust properly. If we don't do that, we will get stuck in the past and won't be able to solve the problems of the future.

sadly, the closer we get to superintelligence, the more paranoid frontier labs will become this also means that loyalty will be valued far more than competence

a small announcement: I'm almost done with a puzzlescript-based benchmark; you can think of it as ARC-AGI-3.5, maybe even ARC-AGI-4; I expect it to be the most reliable way of assessing agents' abstract reasoning skills in a game environments I went through over 300 games, handpicked the most interesting and diverse set of 40, fixed a ton of bugs, went through multiple iterations of prompt design, etc most games can be completed in around 1h, some of them are significantly harder; models are tested using standard harnesses (codex/claude code) with no code generation allowed; action limits are very generous and unfortunately the benchmark is already somewhat saturated, but the models are still insanely action inefficient FYI, I won't release the game set to the public, but I guess I might share runs from some of the rejected games. btw, if anyone knows, what do I have to do in order to not pay $10K out of my own pocket for testing, I'm all ears