OpenAI Opens 719 AI-Generated Math Manuscripts to Review

OpenAI has released 719 manuscripts generated by an unreleased AI model for public examination, giving mathematicians a large collection of claimed results while leaving their correctness and significance subject to review.

The manuscripts are organized into 372 result families. Each family can include a main result, companion arguments, consequences or alternative proofs, so the total does not represent 719 separate problems solved.

The collection spans number theory, algebraic geometry, analysis, theoretical computer science and mathematical physics. The model was given approximately 4,000 problems, and each successful result used an average of three hours of ChatGPT Pro thinking compute.

The repository also includes abridged reasoning summaries for 10 results. The model itself has not been released publicly, limiting independent efforts to reproduce the work.

About 42% of the top-line results have been formalized in Lean. Lean is a proof assistant that checks whether a formal proof logically follows from its stated assumptions. That verification does not by itself establish that a formal statement captures the original research question, that the result is novel or that it is mathematically important.

Some results without Lean formalizations may contain issues. The repository also identifies work that builds on earlier model-generated results and exceptions to the standard production process involving a zero-free region for the Riemann zeta function and the Hodge Conjecture for CM abelian varieties.

The current repository count differs from the 722 manuscripts described in TokenPost’s earlier coverage of OpenAI’s initial release. The reason for the change is not stated.

The Institute for Advanced Study’s Advisory Group on Mathematics and Artificial Intelligence described the publication as the start of a review process. “Making this work public is a first step. This release is the beginning, not the completion, of the process of human understanding and the incorporation of the work into mathematical knowledge,” the group said.

That review will determine which results can be reproduced, formally verified and incorporated into established mathematical research.