OpenAI is releasing a broad range of new mathematical results produced by an internal frontier model. To improve how these results are shared with the math community, the company has consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study and drawn on its advice and public recommendations to shape the release.
For this release, the results are being published in a GitHub repository with protocols for paper revisions and citations. OpenAI says it is still exploring other community-hosted alternatives that meet the committee's guidelines. For future releases, the company commits to improving paper quality through citations, mathematical exposition, and presentation of results. Many of the proofs are also being shared as formalizations in Lean, a programming language that allows a computer to check mathematical proofs, with more to be added as they become available.
To promote transparency, the repository also includes additional details about how the results were obtained. These include 10 summaries of the model's reasoning, estimates of compute spent in terms of ChatGPT Pro usage, and statistics on the number of attempted problems. The average result used the equivalent compute of roughly three hours of ChatGPT Pro thinking.
OpenAI says it hopes this progress will push the frontier of human knowledge. It plans to fund a series of workshops, conferences, and special programs focused on understanding major results produced by AI, and it is working to responsibly release the model behind these results. The company will continue to evaluate its internal frontier models on mathematics and other sciences and to update its standards for disclosing major scientific advancements based on community feedback.
Source: OpenAI Blog · Summarized by HeadlinesBriefing