On October 26, OpenAI said its models had produced three results at the foundations of mathematics, computer science and physics. Three weeks earlier, the company had posted hundreds of AI-written math manuscripts at once and drawn an angry response from mathematicians. This time, its announcement described a new approach. Outside experts advised a different release route for each field. OpenAI released the math output publicly, leaving researchers free to organize their own efforts to explain it. For the computer science and physics results, named outside academics lead working groups that are to check and explain the work. OpenAI says it seeks no academic credit for the discoveries.
Computer scientist Alexander Wissner-Gross, founder of Reified, took a less charitable view on an episode of Moonshots with Peter Diamandis recorded the next day. In his view, OpenAI had enlisted prominent researchers "as almost emissaries, ambassadors." Then he reached for a blunter word: "Meat puppets."
The three claims
Host Peter Diamandis, founder of XPRIZE and Singularity University, laid out the results. In math, he said, OpenAI had proved "a 1991 conjecture about how waves focus in every dimension." OpenAI's announcement calls this a proof of local smoothing in every dimension. Local smoothing says that extreme focusing of waves is short-lived. The company says the proof also implies results on the Kakeya and restriction problems, long-studied questions in the same area of mathematics. On an expert's recommendation, OpenAI released the AI-written proof and its Lean formalization publicly, as it did on October 6, rather than forming a working group around this result.
In computer science, Diamandis said, OpenAI had disproved the exponential time hypothesis, "a widely believed assumption about how hard certain problems are to solve, which hundreds of papers rely on." The hypothesis concerns Boolean satisfiability, the problem of deciding whether a logical formula can be made true. Roughly, it holds that no method can solve this problem dramatically faster than exponential time, meaning the work grows explosively with the size of the formula. OpenAI says its refutation is formalized in Lean, a proof language in which a computer checks every step.
In physics, Diamandis said, OpenAI had found a rigorous proof that quantum chromodynamics (QCD), the theory of the strong force that holds atomic nuclei together, "exists as real mathematics in four dimensions." The announcement describes constructive existence results for four-dimensional quantum field theories, including QCD and pure Yang–Mills theory, building on Balaban's earlier constructive approach. Diamandis said the physics team was led by Nobel laureate David Gross. According to OpenAI, it released the math and computer science artifacts, but Gross's group chose to hold back its physics artifacts at first while it writes the work up.
Diamandis stressed that none of the claims had been peer reviewed. "AI is now producing results at the frontier of human knowledge so fast that humans can't check them in time," he said. "The bottleneck in science has shifted from discovery to verification."
What changed since October 6
The earlier release went out all at once. On October 6, OpenAI posted hundreds of manuscripts produced by an unreleased internal model. The next day, the Association for Human Mathematics issued a statement objecting to the mass publication and urging mathematicians to stop working with OpenAI.
The new announcement handles the results differently. OpenAI calls the model outputs research data that still need interpretation. It says the outside collaborators take part independently, are unpaid and have signed no nondisclosure agreements. The arrangements differ by field. The math result was released publicly, the computer science group published its proof right away, and the physics group withheld its artifacts.
How big are the results?
Diamandis then asked guest Emad Mostaque for his take, and the discussion turned to ranking the results. In that exchange, all three were placed above an earlier result on Riemann zeros and "maybe a little bit smaller than the Navier-Stokes one," a reference to OpenAI's earlier claim about the equations of fluid flow. Some of the work could be checked right away "because they've got the lean," though there was doubt about the write-ups: "Probably the papers are written like slop, but we'll see."
The Kakeya result was put at the level of a Fields Medal, the top prize in mathematics. The bigger surprise was the exponential time hypothesis: "We all thought it was the other way around, that it was true, not false."
The withholding of the physics result drew criticism as "very annoying" and not the right thing to do. There was also "a little bit of drama," the discussion noted, because the physics result extended open-source work that was acknowledged only in footnotes. Earlier in the episode, the conversation had already touched on the physics claim. It was described there as half of the Yang–Mills Millennium Prize problem, and as something OpenAI first gave to string theorists to check, letting them decide when to release it. That approach was criticized as not how science should work.
The talk then widened to credit. "What is academic credit anymore? If you're doing anything that is formally verifiable, does anyone really believe you're not using AI?" The idea that a discovery is special because it "came from my mind as I was hit by an apple" was mocked. Researchers should use the tools to try new approaches, the argument went, rather than stay stuck in the directions that fields such as string theory had pushed them. "Let's try different things, I think is the headline."
Half of Yang–Mills?
Wissner-Gross said he had "called" Yang–Mills when he and Mostaque discussed which Millennium Prize problem, one of a set of famous open problems with prizes attached, was likely to fall next. It was "not quite clear," he said, whether the result covered "some or most or potentially all of Yang Mills."
The discussion pinned that down: the claim was existence, "but not mass gap." The prize problem asks for both, and OpenAI's announcement describes existence results only. Wissner-Gross settled on calling it a "potentially major step towards Yang Mills."
The "meat puppets" argument
Wissner-Gross said OpenAI "seems to have learned some lessons, but not a lot of lessons" from what he called the "slop drop," his name for the October 6 release. He described the reaction to it this way: "mathematicians throwing themselves out of windows, writing petitions, saying it's the end of the field, mass depression, dog sleeping with cats."
OpenAI's answer, in his reading, was to round up mathematicians, physicists and computer scientists. He said he did not want to call it "a veneer or whitewashing," and described the new release as "version 2.0 of a slop drop." The researchers' job, as he sees it, is to make "the bitter lesson of AI go down a little bit more smoothly." The phrase borrows the title of computer scientist Rich Sutton's argument that general methods powered by computation eventually beat approaches built on human expertise. Wissner-Gross described the academics as "an intermediate layer of humans to serve as intellectual ambassadors" who translate results "into something that the community can metabolize more healthfully."
He does not expect the arrangement to last. "I don't think this phase is going to last longer than five minutes," he said, calling it a response to "the moral panic of the slop drop and Navier Stokes." He apologized for his analogy before making it: Hollywood's Motion Picture Association once filtered what the American public could see, most strongly in the McCarthy era. We are now in "a McCarthy-esque era," he said, "where the frontier labs are so terrified of the academic researchers for the moment that they feel the need to clothe themselves in academic researchers to appear legitimate." After the panic, he expects "an abundant solutions phase."
Why Dave Blundin sympathizes with the academics
Dave Blundin, founder of Link Ventures, agreed that "McCarthy-esque is exactly the right phrase," but he had some sympathy for the people on the other side. He said it is "really fun to work with 20, 21, 22-year-olds, because they don't have all these hang-ups." Someone who has worked for 30 years to reach a position in academia where they "get to bless every idea," he said, is "very threatened" when AI works around them. "If I spent 40 years trying to become the head of a department at a major university, I don't want to be irrelevant overnight." The reaction may last only "five minutes in the grand scheme of history," he said, "but it's very, very real."
Blundin sees a pattern in OpenAI's behavior. Its talk about white-collar automation and disappearing jobs created "huge backlash," and it "learned a little bit." The math release brought another backlash, and it learned a little more. He described the new step as OpenAI hiring "the top scientists in the world to be meat puppets and bless things." OpenAI's own account differs on one point: it says the collaborators are unpaid and take part independently. Blundin still accepted the arrangement for now. "Okay," he said, "I guess that's okay in transition."
Listen to the full discussion on the episode page.