🔍 Read the full analysis: What’s Next For OpenAI’s AI Mathematics After 722 Proofs? on ThorstenMeyerAI.com
Get tech for your team delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
OpenAI published 722 mathematical manuscripts produced by an unnamed, unreleased model, covering results selected from about 4,000 problems. The work includes extraordinary claims, but outside mathematicians have not yet verified the collection; its significance will depend on whether the proofs hold and researchers can understand and use them.
OpenAI published 722 mathematical manuscripts on Monday, attributing them to an unnamed, unreleased model that worked on roughly 4,000 problems. The collection includes claims involving long-standing open problems, but OpenAI chief executive Sam Altman said they have not been confirmed by outside mathematicians, leaving their validity and potential impact unresolved.
The manuscripts are arranged into 372 families of related results across fields including number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. OpenAI says the model spent an average of about three hours of ChatGPT Pro thinking compute on each result. The files are published under the Apache-2.0 license, and the project repository includes Lean formalizations for many, but not all, of the claims.
Among the reported results are a proposed proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, and a claim about the isomorphism of nonabelian free group factors. Other manuscripts concern a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12, the Hodge conjecture for CM abelian varieties and conjectures in convex geometry. These are claims in the released material, not independently established discoveries.
OpenAI’s repository cautions that some unformalized results could have issues. The company also supplied ten abridged reasoning summaries, covering only a fraction of the 372 families. According to the source report, OpenAI selected the problems and filtered them for what it considered an appropriate level of significance; no outside group made that selection. The Riemann write-up and the Hodge result were exceptions to the usual process, with the Riemann manuscript edited by humans for readability.
722 proofs, one question: will any of OpenAI’s AI mathematics actually lead anywhere?
An unreleased, unnamed model produced claimed proofs of results that would each define a career. Sam Altman calls them “claims not yet confirmed by outside mathematicians.” The real question isn’t whether it’s impressive. It’s whether answers nobody understands become discoveries anyone can build on.
Same day: Alon, Bloom, Gowers, Litt, Sawin post a digested, human-verified version. The model for success.
Connes rigidity counterexample challenged within a day — constructed groups fail the required condition. Three rival machine “counterexamples” from different labs now circulate.
~10,000 agents, 88 hours, est. ~$22M at retail. Priority dispute; 25 Fields Medalists sign “A Severe Misalignment” — not saying it’s wrong, saying it’s not understood.
Altman now hedges at announcement — a shift from September. Verification has barely started.
Humans extract the technique, write it up, build on it. This is where downstream discovery comes from.
The question is answered; nobody learns anything reusable. Closes a door without opening a field.
The proof breaks, or proves a statement that doesn’t match the conjecture as mathematicians mean it.
The Unique Games Conjecture is the clearest case. Results like the optimality of Goemans–Williamson for Max-Cut are proved assuming UGC. A correct proof converts them all — no understanding required. A zero-free strip for zeta works the same way for prime-distribution results. Free group factors, Kadison, Mahler would redirect whole programmes — but how depends on the method, which means digestion.
Technology. A Navier–Stokes blow-up proof doesn’t change how anyone designs aircraft; engineering turbulence models never depended on the answer. Near-term consequences are mathematical, not industrial. “AI will cure cancer next” skips several steps.
“Verification abundance, adjudication scarcity” — making proof-checking cheap doesn’t reduce the burden of deciding what’s true and what matters. 722 manuscripts land on a review system built for a trickle, filtered by a selection nobody outside OpenAI made.
Humans re-deriving results, like Alon–Gowers et al. in May
Other people’s work building on these manuscripts
How many unformalized results survive expert checking
Do the Lean statements match the real conjectures?
Do any survive peer review?
Some of it, yes — where a literature is waiting (UGC), a correct proof pays off immediately; where a proof carries a new technique humans digest, it can open a field. Most of it, probably not on its own: at 722 manuscripts with 10 reasoning summaries, the Four Colour pattern is the likely default unless mathematicians are funded and given time. And some will be wrong — OpenAI says so itself. It’s an industry pattern, not one company’s: the forced-Euler result came from an Anthropic researcher, and rival machine-generated Connes “counterexamples” circulate from different labs. The proofs arrived this week. The discoveries, if they come, will arrive at the speed of human understanding.
Why Verification Will Shape the Impact
The release matters because the claims, if correct, could affect research well beyond the individual problems. The Unique Games Conjecture, for example, underpins a substantial body of theoretical computer science about the limits of approximation algorithms. A proof could prompt researchers to revisit results that rely on the conjecture. That consequence is conditional: mathematicians must first establish that the argument is sound and proves the relevant statement.
For mathematics, a correct answer is not always the most valuable outcome. A proof can matter because its methods become tools that other researchers adapt. The key test for this collection is whether people can extract understandable, reusable ideas, rather than merely check a conclusion or accept a machine-generated argument. Formal verification can help establish that a proof follows from stated assumptions, but it does not by itself show that the result is important, explanatory or useful to other fields.
The release also sharpens a debate about what AI progress in mathematics should mean. A system may solve a famous benchmark problem without producing insight people can build on. For working mathematicians and computer scientists, the practical question is whether these manuscripts save time, generate methods or redirect research—not simply how many claims they contain.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Math Releases
This is OpenAI’s fourth major mathematics release this year, according to the source report. The earlier releases offer examples of both the potential and the risks of evaluating machine-generated work before independent review is complete.
In May, OpenAI reported that its model had found a counterexample to the Erdős unit-distance conjecture, dating to 1946. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—then published what they called a digested, human-verified version. That process gave researchers a form they could assess and discuss, rather than requiring them to rely on raw model output.
In August, the company announced ten advances. One claimed counterexample to Connes’s rigidity conjecture was challenged within a day: a critique said the constructed groups did not meet the condition required by the conjecture. In September, OpenAI announced a Lean-formalized result on finite-time blow-up for the Navier–Stokes equations, produced, the company said, by about 10,000 concurrent agents over 88 hours. That announcement also drew a dispute over priority and a declaration signed by 25 Fields Medalists criticizing the use of famous problems as benchmarks without human understanding. The disagreement was about the purpose and presentation of AI work, not a shared verdict that the proof was wrong.
“Digested, human-verified.”
— Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin, describing their May work
formal proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Which Claims Will Survive Review
The central unknown is whether the manuscripts’ arguments are correct and match the claims they are presented as proving. Independent mathematicians have not confirmed the collection, and formalizations exist for many, not all, of the results. The repository’s warning about unformalized work makes clear that the release is not a blanket certification.
It is also unclear how much of the work can be independently checked in a useful timeframe. OpenAI released ten abridged reasoning summaries for 372 families, and the source report says the company controlled which problems and results were selected for publication. The criteria behind that selection, and how the unrepresented families compare with the showcased claims, are not established in the supplied material.
Even if particular proofs are correct, their longer-term value remains open. Researchers may find new techniques, confirm a result without learning a reusable method, or identify errors or mismatches. The collection’s ultimate significance cannot be inferred from the number of manuscripts or the prominence of the problems alone.
AI-assisted mathematical research tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks and Usable Proofs
The immediate next step is independent mathematical scrutiny. Researchers will need to examine individual manuscripts, check formalized arguments where available and translate complex reasoning into forms that specialists can evaluate. The May Erdős result provides one possible model: mathematicians prepared a digested version and verified it. OpenAI has not, in the supplied material, provided a timetable for outside review of all 372 families.
Readers should watch for specific, attributable follow-up: named mathematicians or research groups reporting that a result has been checked, corrections or counterarguments, and explanations of what a proof contributes beyond the conclusion. A formalization or a positive review of one manuscript would not validate the entire collection; each claim requires its own assessment.
The longer-term measure will be whether researchers can build on the work. If they extract techniques, apply them elsewhere or use verified results to settle dependent questions, the release could have lasting value. If proofs are difficult to interpret, fail review or settle claims without offering reusable insight, the collection’s impact may be narrower than its headline claims suggest. For now, the confirmed development is the publication itself; the mathematical status of its most ambitious results remains unsettled.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, organized into 372 families and attributed to an unnamed, unreleased model. The company says the results came from roughly 4,000 problems posed to the model.
Have mathematicians verified the claims?
Not as a collection. The source material says the claims have not yet been confirmed by outside mathematicians, and OpenAI’s repository warns that some unformalized results could have issues. Individual manuscripts may receive separate review.
Does the release prove the Unique Games Conjecture?
OpenAI’s collection includes a manuscript claiming a proof of the Unique Games Conjecture. The claim has not been independently confirmed in the supplied material, so it should not be treated as a settled result.
Why does it matter if a proof is correct but hard to understand?
A correct proof can settle a question, but mathematics often advances when researchers understand and reuse the methods behind it. Whether the manuscripts produce transferable ideas, rather than conclusions alone, will help determine their broader value.
What should happen next?
Mathematicians need to review the manuscripts individually, check formalized proofs where available and publish clear accounts of what holds up. The next meaningful milestones are independent verification, corrections or rebuttals, and evidence that researchers can use the methods.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
