LEAN

OpenAI put hundreds of model-written math manuscripts on GitHub, about 42% of top-line results with Lean proofs. What the repo actually claims, how the Lean check works, where Hacker News pushed back, and the verification lesson for anyone shipping agent output.

Anthropic says Claude worked largely autonomously for 11 days to formalize Fermat's Last Theorem in Lean. The developer lesson is less about one theorem and more about verified repo-scale research artifacts.

OpenAI's next model, codenamed Astra, produced results on ten problems open for at least a decade - including non-sofic groups and Erdős problems 146, 180, and 183 - with every argument formalized as a Lean certificate.