OpenAI announces three big science discoveries and a new way to work with academics
OpenAI says its AI systems have produced three significant results in mathematics, computer science and physics, and is involving outside academics in examining some of them. The announcement matters because the claims concern long-standing or widely accepted problems, while also raising questions about how AI-generated research should be checked, shared and credited.
In mathematics, OpenAI released a proof concerning local smoothing for the Euclidean wave equation, with implications for the Kakeya conjecture and Stein’s spherical restriction estimates. In computer science, it claims a deterministic subexponential-time algorithm for counting satisfying assignments that would disprove the Exponential Time Hypothesis; three researchers will examine the proof, which is available with a Lean proof on GitHub. The physics result, concerning the existence of certain four-dimensional quantum field theories including QCD, has not yet been fully published, and will be reviewed by a group led by David Gross. OpenAI’s advisers have cautioned against giving selected researchers early access, citing risks of unequal access and biased selection.
In short
- OpenAI reports new AI results in maths, computer science and physics.
- A computer science claim would overturn the Exponential Time Hypothesis.
- Academics will examine proofs as debate grows over access and credit.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
AI systems may help researchers tackle problems that have resisted conventional approaches, and these claims could be valuable if their proofs withstand scrutiny. Publishing the computer science proof with a Lean formalisation and inviting independent academics to review results are reasonable steps towards verification, while collaboration with specialists may help assess work that is difficult to evaluate quickly.
The case against
Claims about major results should be treated cautiously until the methods and evidence are available for broad, independent assessment; one result is not yet fully published, and review by selected academics may not provide enough transparency. Early access can also give a small group undue influence over scrutiny and credit, while making it harder for other researchers to check the work on equal terms.
Explained simply
Mathematics, computer science and physics all contain problems whose solutions can affect many other results. A proof is a step-by-step argument that experts check; in some areas, computer programs can also help verify that each step follows.
OpenAI develops artificial intelligence systems, and says some of its systems have produced results in these fields. The claims involve questions about wave equations, the time needed to solve certain computing problems, and the mathematical foundations of quantum physics, including the theory used to describe quarks and gluons.
Outside academics are being asked to examine some of the work, but the checking and publication arrangements differ between the claims. How researchers verify, share and credit AI-assisted work matters to the wider scientific community, especially when some people may see results before others.