Sharing AI progress in mathematics
OpenAI releases new mathematical results generated by an internal frontier model, sharing proofs, reasoning summaries, and compute details via GitHub and formalized in Lean.
OpenAI has published a collection of new mathematical results produced by one of its internal frontier models, marking a step toward greater transparency in AI-driven research. The company consulted with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study to establish best practices for sharing such results. The release includes protocols for paper revisions and citations, with plans to explore additional community-hosted platforms that align with the committee’s guidelines.
The results are shared through a GitHub repository, which also features formalized proofs in Lean, a programming language designed to verify mathematical proofs computationally. OpenAI intends to expand the repository with more formalized proofs as they become available, aiming to enhance clarity and reproducibility in mathematical research facilitated by AI.
To provide deeper insight into the model’s performance, OpenAI included supplementary details in the repository, such as 10 summaries of the model’s reasoning process, estimates of compute usage in terms of ChatGPT Pro usage, and statistics on the number of attempted problems. The average result required compute equivalent to roughly three hours of ChatGPT Pro thinking.
The company emphasized its commitment to scientific transparency and announced plans to fund workshops, conferences, and special programs focused on understanding major AI-generated mathematical results. OpenAI also reiterated its goal of responsibly releasing the model behind these findings and continuing to evaluate its frontier models to advance scientific research.