OpenAI Says Its Models Produced Three New Results in Math, Computer Science and Physics

The three outputs include a Lean-formalized disproof of the Exponential Time Hypothesis for 3-SAT, a proof of the local smoothing conjecture that would also settle Kakeya and Stein's restriction conjectures in all dimensions, and a physics result that contains a proof of pure Yang-Mills existence, a piece of a Clay Millennium problem. OpenAI describes the outputs as raw research data that outside experts must turn into understanding. It asked external shepherds to organize groups under no-NDA terms, claims no academic credit for the discoveries, and says participation is voluntary and unpaid. Three release strategies were proposed, and OpenAI admits it does not yet know which will work best. For founders, the process matters as much as the proofs. A lab is treating model output as a dataset and asking humans to supply interpretation and recognition, which is a model other AI-for-science startups may copy. Formal verification was the step that made these claims checkable, so teams selling research tools should plan for it.