OpenAI unveiled three major research discoveries across mathematics, computer science, and physics less than a month after releasing hundreds of mathematical manuscripts, adopting a new verification model that deploys external experts to validate results.
The shift represents a response to criticism over the company's previous release methodology. For the computer science and physics findings, OpenAI will use external experts as "shepherds" to guide verification—a departure from direct release that faced scrutiny over access and validation rigor.
In mathematics, one discovery addresses "Local smoothing for the Euclidean wave equation," establishing Sogge's local smoothing for waves in spatial dimensions of at least two, with findings that yield the Kakeya maximal and dimension conjectures alongside Stein's spherical restriction estimates. OpenAI is releasing the raw model output directly, allowing independent researchers to develop their own explanations.
For computer science, OpenAI claims to have disproven the Exponential Time Hypothesis (ETH)—a foundational assumption in computational complexity. The paper presents a deterministic algorithm capable of exactly counting satisfying assignments of fixed-width Conjunctive Normal Form formulas in subexponential time and polynomial space, effectively refuting ETH for 3-SAT.
Verification will be overseen by an independent working group including Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams. OpenAI released all relevant materials immediately at the group's request.
The physics discovery—a "mathematically rigorous existence proof of certain interacting quantum field theories containing gauge fields and fermions in four dimensions"—remains unreleased pending verification. This includes Quantum Chromodynamics and potential unified completion of the Standard Model of particle physics. A verification group led by David Gross, with Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov, will oversee the work.
OpenAI acknowledged the collaborative approach introduces new risks. The company's own AGMAI panel of mathematics advisers recently recommended against practices designating specific individuals to receive AI output first, citing concerns about unequal access, reputation borrowing, and biased selection. The panel instead recommended AI laboratories release outputs on scholarly repositories, include Lean-formalized proofs, disclose problem selection processes, and fund research to validate results.
