OpenAI posts 722 AI-generated math papers on GitHub

OpenAI's exercise reaches across mathematics well beyond a single headline result. The papers, posted to GitHub, touch about 20 subfields, with theoretical computer science alone accounting for more than 80 of them. Three of those papers address matrix multiplications, the operations AI models use to process data, and the model developed a clearer definition of the limit on how much that operation can be sped up. A separate paper offers a new algorithm for multiplying integers, another foundational building block of many programs.

The physics-adjacent results are notable in their own right. More than a dozen proofs relate to partial differential equations, which are essential to chip design, architecture, quantum mechanics and other fields. The model proved a version of De Giorgi's conjecture, tied to an equation used to study metal alloys, and clarified questions around the Navier–Stokes equations, which engineers use to study how liquids flow. In September, the same model solved a different Navier–Stokes problem that ranked as one of the most difficult open questions in mathematics.

The stakes likely hinge on verification. Many papers include Lean files, which let a computer quickly check a new proof, and OpenAI says it plans to release Lean proofs for more of the papers and to finance research events and programs focused on reviewing AI-generated math discoveries. Whether these results hold up under that scrutiny is what will determine how much weight mathematicians give them.