OpenAI posts new math results from internal model in GitHub repository

OpenAI frames this release as an attempt to share machine-produced mathematics in a form the math community can actually work with. It says it consulted with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study while developing best practices, and drew on that advice and public recommendations for how the results are presented. The choice of a GitHub repository, with protocols for paper revisions and citations, places the results in a venue where they can be inspected, cited, and revised over time, and OpenAI says it is still exploring other community-hosted alternatives that meet the committee's guidelines.

The repository also carries material aimed at transparency: formalizations of many proofs in Lean, a programming language that lets a computer check mathematical proofs, plus 10 summaries of the model's reasoning, estimations of compute spent in terms of Pro usage on ChatGPT, and statistics about attempted problems. OpenAI states that the average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking.

Looking ahead, OpenAI says it will fund workshops, conferences, and special programs around understanding major results produced by AI, and that it is working to responsibly release the model that produced these results. Whether the release functions as a durable bridge to the math community is likely to hinge on how the promised future improvements in citations, exposition, and presentation are received, and on whether the community-hosted alternatives the company says it is exploring materialize.