OpenAI announced three results on Monday in mathematics, computer science and physics, along with what it calls an evolving approach to sharing its findings. Researchers who spoke to The Verge said the science looks significant, but many were frustrated by how the company is handling its release.
The announcement follows a tense few months between OpenAI and mathematicians, who are still dealing with the hundreds of preprints the company released at the start of October. This time, OpenAI says it is working with expert "shepherds" to decide how each result should be shared. Each of the three has a different strategy. The mathematics output was published, leaving researchers free to organize their own efforts to explain it. For the computer science finding, OpenAI will assemble an external group to explain the result. The physics work has not been released, and OpenAI said it will defer to outside experts on how to do so.
The mathematics result purports to resolve the Kakeya conjecture in all dimensions. Roughly, the problem asks how little space is needed to rotate a needle in every direction. A proof of the three-dimensional version was among the achievements for which NYU mathematician Hong Wang received a Fields Medal earlier this year. Brown University professor Brendan Hassett said the claim "looks well supported," though he had not yet checked the Lean formalization, a computational method of verifying results.
The physics result concerns quantum chromodynamics, which describes the strong nuclear force, and its relationship to Yang-Mills theory. That theory sits at the heart of one of the seven Millennium Prize problems, although OpenAI's work does not address the unresolved "mass gap." The computer science result claims to disprove the exponential time hypothesis, a widely held assumption that certain problems cannot be solved much faster than by brute force. It is closely related to P versus NP, another Millennium Prize problem.
Experts described the results as "huge," "major breakthroughs" and "significant," but fully judging them will take time. Many researchers questioned why OpenAI is adding barriers by choosing who oversees scrutiny. Constantin Kogler of the Institute for Advanced Study welcomed the smaller batch of three results, but said, "It doesn't make any sense that first only a few distinguished scientists can look at them."
Andreas Karch, a physics professor at the University of Texas at Austin, was more blunt. "The fact that they handpick the group is a disaster," he said. "The large labs have enough power as it is. They should not be picking winners and losers." He asked why OpenAI "can't release the proof now" while also appointing its experts, arguing this "would democratize access while still showing that they care about making the results meaningful and understandable."
The Advisory Group on Mathematics and Artificial Intelligence, a panel OpenAI had pledged to consult, warned against this kind of practice a day before the announcement. It said selecting people to receive output first is "unnecessary," borrows the reputation of chosen researchers and risks unequal access, biased selection and delays. It recommended releasing output on scholarly repositories with formalized proofs, and funding people who work on understanding the results. After the announcement, the group urged OpenAI to publish the physics outputs without delay, noting the Lean formalization is now complete.
Hassett said open availability could also support faster adoption of new tools. Stanford mathematician Jared Duker Lichtman was more resigned: "for any approach, there is going to be someone who is upset," he said. Xiaoyu He of the Georgia Institute of Technology called the results very impressive, but said OpenAI is making a show of responsibility without addressing the main issues: that fundamental results are being solved so quickly without human contribution, and that the papers are hard to read and verify.
