OpenAI publishes 722 math manuscripts from an unreleased internal model

OpenAI has released a large batch of new mathematical results produced by an internal frontier model it has not released to the public.

722 manuscripts went up on GitHub. On October 6, OpenAI published 722 manuscripts grouped into 372 result families in a public GitHub repository, with PDFs, source files and protocols for revisions and citations. A family groups related papers, such as a principal result, companion arguments, consequences or alternative proofs.

Not everything has been machine-checked. OpenAI is sharing Lean formalizations, which let a computer check a proof, for many of the results, but not all of them. The repository says some of the unformalized results could have issues, and OpenAI says it will add more formalizations as it obtains them.

OpenAI disclosed how the results were made. The repository includes 10 abridged summaries of the model's reasoning, compute estimates and problem statistics. OpenAI says the model was posed about 4,000 problems over the evaluation, and that each result used roughly three hours of ChatGPT Pro thinking compute on average.

Mathematicians were consulted on the release. OpenAI says it worked with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study and followed its recommendations. It is exploring community-hosted alternatives to GitHub and says future releases will improve citations, exposition and presentation.

More is planned. OpenAI says it will fund workshops, conferences and special programs around major AI-produced results, and is working to responsibly release the model that produced them.