Today is a big day, ARC Prize version 2 has just been released! Watch the launch video on MLST.

Today we are announcing ARC-AGI-2, an unsaturated frontier AGI benchmark that challenges AI reasoning systems (same relative ease for humans). Grand Prize: 85%, ~$0.42/task efficiency Current Performance: * Base LLMs: 0% * Reasoning Systems: <4%

👀

A TikTok-style clip featuring a repeatedly watched moment from a movie or series on Netflix, with a hook, would be a banger for discoverability.

o3 and o4-mini on ARC-AGI's Semi Private Evaluation * o3-medium scores 53% on ARC-AGI-1 * o4-mini shows state-of-the-art efficiency * ARC-AGI-2 remains virtually unsolved (<3%) Through analysis we highlight differences from o3-preview and other model behavior

...evolutionary agents

We put years of work in on this hard problem and made progress. We're seeking new breakthroughs for ARC-AGI-2.

Alright, time to read what all the hype is about.

Here are all the architecture tricks used by gpt-oss: - Attention sinks - for each attention head, have a learned scalar such that softmax(qk) becomes softmax over [a_1, a_2, ..., a_T, sink]. Tokens don't have to attend to anything if all the attention scores are low! -

GPT-5 will power a new web-based PDF reader for research papers I'm building.

OK, it is a bit underwhelming after playing around for a while.

I've written the full story of Attention Sinks — a technical deep-dive into how the mechanism was developed and how our research ended up being used in OpenAI's new OSS models. For those interested in the details: hanlab.mit.edu/blog/streaming…

Can’t help but feel the crushing weight of diminishing returns. We need a new S curve.

AI explanations + visualization tightly integrated. Understand complex concepts intuitively, in minutes. Can't wait to ship