OpenAI releases three AI-generated breakthroughs and new academic collaboration model

OpenAI reported three AI-driven discoveries across mathematics, computer science and physics. In mathematics, the system produced a proof of the local smoothing conjecture, which also resolves the Kakeya and Stein restriction conjectures in all dimensions. In computer science, it presented a deterministic algorithm that counts satisfying assignments of fixed-width CNF formulas in subexponential time, thereby disproving the Exponential Time Hypothesis, with a verification team led by Russell Impagliazzo, Ramamohan Paturi and Ryan Williams.

The physics claim is a rigorous existence proof for interacting quantum field theories that include gauge fields, fermions, QCD and a unified Standard Model, shepherded by David Gross and colleagues Michael Douglas, Arthur Jaffe, Igor Klebanov and Nikita Nekrasov. OpenAI released the AI output and Lean proofs on GitHub and issued an open letter asking the community to discuss early-access norms, credit allocation, and support for researchers as AI-generated results become more common. The company also noted criticism from its own AGMAI advisory panel regarding selective early access and recommended broader open-repository releases in the future.