OpenAI claims three scientific discoveries, pilots academic shepherds for release
OpenAI says internal models produced results in harmonic analysis (local smoothing/Kakeya), complexity theory (claiming the Exponential Time Hypothesis is false), and quantum field theory (existence proofs including QCD), and is testing three expert-advised release strategies. It says the CS natural-language proof and Lean formalization are public on GitHub, while the physics group is deferring AI artifacts. Treat discoveries as OpenAI claims; independent verification is ongoing.
OpenAI announced three discoveries in mathematics, computer science, and physics—about harmonic analysis, complexity theory, and quantum field theory—and frames the manuscripts as raw research data. External experts were asked to shepherd working groups that turn those outputs into scientific understanding, advising three strategies: release the math AI output for the community to organize explanations; form an external CS collaboration while publishing the AI output; form a physics collaboration and defer to it on releasing materials. Stated principles: transparency (no NDAs; free discussion and publication), no academic credit sought for discoveries from internal models, recognition for the groups’ understanding and write-ups, and voluntary unpaid independent participation.
In math, OpenAI claims a proof of Sogge’s local smoothing conjecture in every dimension (implying Kakeya and Stein restriction in all dimensions) via a novel auxiliary PDE; AI output dated 2026-10-18, Lean the next day, with NL write-up and Lean proof on public GitHub. In CS, it claims a Lean-verified refutation of the Exponential Time Hypothesis, to be shepherded by Russell Impagliazzo, Ramamohan Paturi (UC San Diego), and Ryan Williams (MIT), with raw data public and a funded conference; NL disproof 2026-10-16, Lean 2026-10-17, also released. In physics, it claims rigorous existence proofs for certain interacting 4D QFTs with gauge fields and fermions, including QCD and a unified Standard Model completion, plus pure Yang–Mills existence (related to the Clay problem); David Gross will lead a group including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov; NL Yang–Mills 2026-09-09, QCD 2026-09-18, Yang–Mills Lean finished 2026-10-16; the physics group is holding AI artifacts and writing them up itself. OpenAI also notes a recent $5 million contribution to the Simons Foundation. All discovery claims are OpenAI’s; independent verification is ongoing.