OpenAI released a collection of 722 mathematical manuscripts on GitHub, detailing solutions and progress on hundreds of long-standing problems. The papers, grouped into 372 distinct result families, were produced by an internal AI model that OpenAI has not yet made public. This release follows OpenAI's earlier announcement of a claimed solution to the Navier-Stokes existence and smoothness problem.
The newly published works span various mathematical fields, including number theory, algebraic geometry, theoretical computer science, and differential equations. Among the claimed breakthroughs are solutions to the four-dimensional Kakeya conjecture and progress toward the Riemann hypothesis. OpenAI stated that the average result required approximately three hours of computation equivalent to ChatGPT Pro usage. The company also provided Lean formalizations for many of the proofs, a programming language designed to verify mathematical arguments, alongside summaries of the model's reasoning.
The release has generated a mixed reaction from the mathematical community. While the scale and nature of the AI-generated results have impressed some, others have expressed concern regarding research ethics, attribution, and academic conduct. OpenAI established the Advisory Group on Mathematics and Artificial Intelligence (AGMAI) to help guide the responsible communication of its AI-driven mathematical research. However, AGMAI has stated that its advice does not constitute an endorsement of OpenAI's methods or results, emphasizing that this release represents the beginning of human understanding rather than a completion.
Mathematicians have raised questions about the rigor and validation of these AI-generated proofs, particularly given that not all manuscripts are accompanied by formal Lean verifications. OpenAI acknowledged that some unformalized results might contain issues and stated its intention to address them. The company also indicated it would fund workshops and conferences to discuss AI-generated mathematical findings. The unreleased nature of the model that produced these results has also been a point of contention, with some mathematicians urging for greater transparency in AI research.