Mathematical manuscripts and supporting proof artifacts produced by OpenAI

What happened: OpenAI published a repository of mathematical manuscripts and proof artifacts produced by an internal model, covering results at different stages of verification. Key numbers: 719 manuscripts in 372 families; about 42% of top-line results formalized in Lean; roughly 4,000 problems posed; average three hours of ChatGPT Pro thinking compute per result. Why it matters: OpenAI says some unformalized results could have issues and it will fix them, and it is exploring community-hosted repositories. Corrections will be recorded as new versions while prior versions stay accessible.

OpenAI published a repository of mathematical manuscripts and proof artifacts produced by an internal model, covering results at different stages of verification.

OpenAI says some unformalized results could have issues and it will fix them, and it is exploring community-hosted repositories. Corrections will be recorded as new versions while prior versions stay accessible.