OpenAI's science breakthroughs draw researcher backlash — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI announced three results on Monday — a purported resolution of the Kakeya conjecture in all dimensions, a claimed disproof of the exponential time hypothesis closely related to P versus NP, and a physics result on quantum chromodynamics and Yang-Mills theory that does not address the unresolved mass gap.
- OpenAI gave each finding a different release path: the AI-generated mathematics output is public, the company will assemble an external group of researchers to explain the computer science result, and the physics materials have not been released at all, with timing left to an external group of experts.
- Brendan Hassett (Brown University) said OpenAI's Kakeya claim "looks well supported," though he has not checked the Lean formalization, and said many groups worldwide wrestling with the results would support rapid adoption of new tools.
- Constantin Kogler (Institute for Advanced Study) said he welcomed the smaller batch of three results after hundreds of preprints were dumped on mathematicians in October, but that restricting first access to OpenAI's chosen experts "doesn't make any sense."
- Andreas Karch (University of Texas at Austin) called OpenAI's handpicking "a disaster" — large labs "should not be picking winners and losers" — and said releasing the proof now would democratize access while still letting experts explain it.
- AGMAI warned a day before the announcement that AI labs should not give selected researchers first access, citing unequal access, biased selection and delays, and after publication urged OpenAI to release the physics natural-language and Lean outputs without delay since the formalization is complete.
- Xiaoyu He (Georgia Institute of Technology) called the results impressive but said OpenAI is "making a show" of trying to be responsible: fundamental results are being solved so quickly without human contribution, and the papers are difficult to read and verify.
Why it matters: AGMAI — the group OpenAI had pledged to consult — warned a day before the announcement against giving selected scientists first access, citing unequal access, biased selection and delays. With the physics output still unreleased even though its Lean formalization is complete, the outside experts meant to scrutinize it start from less material than the public can see.
Ask SkimNews
