OpenAI claims three science breakthroughs, and experts push back on who sees them first

On Monday, OpenAI announced three scientific breakthroughs in mathematics, computer science and physics. Outside experts say the results deserve serious attention. But several are frustrated that OpenAI is picking who gets to examine some of the work first, and they say claims this big should be open for anyone to check.

What OpenAI announced

The math result claims to resolve the Kakeya conjecture in all dimensions. Roughly, it asks how little space is needed to rotate a needle in every direction. The three-dimensional version was part of the work for which NYU mathematician Hong Wang won a Fields Medal this year.

The computer science result claims to disprove the exponential time hypothesis, a widely held assumption about how fast certain problems can be solved. The physics result concerns the strong nuclear force and a theory called Yang-Mills. It does not address the unresolved "mass gap" problem.

Three results, three release plans

• Math: OpenAI released the AI-generated output so researchers can organize their own efforts to explain it.
• Computer science: OpenAI will assemble an external group of researchers to explain the result.
• Physics: nothing has been released. OpenAI says it will defer to outside experts on how to share it.

OpenAI says it worked with expert "shepherds" to choose these approaches. The release follows a rough few months. Mathematicians are still dealing with hundreds of preprints that OpenAI dumped on them at the start of October.

What experts are saying

Reactions to the science were positive. Experts used words like "huge," "major breakthroughs" and "significant." Brown University's Brendan Hassett said the math claim "looks well supported," but he has not yet checked the Lean formalization, a way of verifying results by computer.

The complaints are about access. Constantin Kogler of the Institute for Advanced Study said he is fine with a small batch of three results. But he said it makes no sense for only a few distinguished scientists to look first. Physicist Andreas Karch of the University of Texas at Austin called the handpicked group "a disaster" and said large labs "should not be picking winners and losers."

A group OpenAI had pledged to consult, the Advisory Group on Mathematics and Artificial Intelligence, had warned against this a day earlier. It recommended posting results on scholarly repositories with formal proofs. After the announcement, it urged OpenAI to publish the physics outputs without delay.

What is not known yet

No one has fully judged these findings. Even with every document in hand, that would take time, and the physics and computer science details are not public. Until outsiders can check the work openly, the claims remain claims, however promising.