🔍 Read the full analysis: What Do 722 Proofs Mean For OpenAI’s AI Mathematics? on ThorstenMeyerAI.com
Get smart everyday buys 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 has published 722 mathematical manuscripts generated by an unnamed, unreleased model, covering 372 families of results from roughly 4,000 problems. The collection includes claims about major open problems, but OpenAI says they have not been confirmed by outside mathematicians, and its repository warns that some results without formal proofs may contain issues.
OpenAI published 722 mathematical manuscripts on Monday, presenting work attributed to an unnamed model that the company has not released. The papers cover 372 families of results drawn from roughly 4,000 problems; claims include solutions to several major open questions, but outside mathematicians have not yet confirmed them.
OpenAI’s repository groups the manuscripts across fields including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. The company says each result took about three hours of ChatGPT Pro thinking compute on average. It selected the published work from approximately 4,000 problems, using its own judgment of which results had “an appropriate level of significance.” The source material says no outside group made that selection.
The catalogue includes claims concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, the Hodge conjecture for CM abelian varieties and conjectures in convex geometry. Another manuscript claims a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. These are claims in the published manuscripts, not independently established solutions.
OpenAI released the work under the Apache-2.0 license. Many, but not all, results have Lean formalizations, a computer-checkable representation of mathematical proof. The repository itself cautions that “some of the unformalized results could have issues.” Only ten abridged reasoning summaries were included for the 372 families. The Riemann manuscript was edited by humans for readability, according to the source material; the Hodge result and the Riemann result were exceptions to the standard process.
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.
Verification Will Determine the Value
The scale and ambition of the claims make independent checking the immediate issue. A correct proof of a major conjecture could alter research in its field, but publication by a company does not establish that a proof is sound. The manuscripts must be checked for gaps, for whether they prove the stated result, and for whether other mathematicians can understand and use their methods.
That last point matters because a proof can settle a question without giving researchers useful ideas. The source material contrasts results that humans can digest and build on with computer-assisted work that answers a problem but yields little reusable theory. It also notes a recent case in which a claimed counterexample to Connes’s rigidity conjecture was disputed because the groups constructed did not meet the conjecture’s required condition. The key measure is not the number of manuscripts, but what survives scrutiny and what knowledge can be extracted from them.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Mathematics Releases
The 722-paper release follows three earlier mathematics announcements by OpenAI this year, according to the source material. In May, a model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians then posted what they described as a digested, human-verified version, illustrating how machine output can be made accessible for review and checked by specialists.
OpenAI’s August “Ten Advances” release drew a mixed response. One claimed counterexample to Connes’s rigidity conjecture was challenged within a day, with critics arguing that the construction did not satisfy the necessary conditions. In September, OpenAI announced a Lean-formalized result on finite-time blow-up for the Navier–Stokes equations, produced using about 10,000 concurrent agents over 88 hours, as described in the source material.
After that announcement, 25 Fields Medalists signed a declaration titled “A Severe Misalignment of AI in Mathematics.” Their concern, as summarized in the source material, was not that the Navier–Stokes proof was wrong, but that treating famous problems as benchmarks without human understanding could conflict with mathematics’ aims. That debate is relevant to the new catalogue, which raises the same question at a much larger scale: whether automated results can become understandable mathematics.
As an affiliate, we earn on qualifying purchases.
Which Manuscripts Will Withstand Review
Independent verification remains outstanding. The source material does not establish that outside mathematicians have confirmed any of the headline claims in the new release. It also does not provide a public accounting of how many of the 722 manuscripts have been checked, how long review will take, or whether the claims will be revised or withdrawn.
The level of formal checking varies: Lean formalizations are available for many results but not all, and the repository warns about possible problems in unformalized work. A formalization can help verify a proof, but the source material does not say that every formalized manuscript has received independent mathematical review. It is also unclear how many of the 372 result families will yield reusable techniques rather than a result that is difficult for researchers to interpret.
formal proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Mathematicians Begin the Proof Checks
The next step is for mathematicians to examine the manuscripts, reproduce the arguments where possible and determine whether each paper proves the claim it makes. Work that is difficult to follow may need to be rewritten or “digested” into a form specialists can assess. The earlier Erdős release offers one example of that process, but it does not establish that the new catalogue will receive the same outcome.
For now, readers should treat the papers as published claims awaiting review, not as settled solutions. The source material gives no confirmed timetable for outside assessments or for OpenAI to report corrections. The standing question is which results withstand scrutiny, and whether any lead to methods other researchers can build on.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
It published 722 mathematical manuscripts, organized into 372 families of related results and attributed to an unnamed model that has not been released.
Are the claimed solutions confirmed?
No outside confirmation is established in the source material. OpenAI described the results as claims not yet confirmed by outside mathematicians, and its repository warns that some unformalized results may have issues.
What does a Lean formalization do?
Lean is a proof assistant that lets a mathematical argument be represented in a form a computer can check. Many, but not all, results in the catalogue have Lean formalizations; that does not mean every result has completed independent review.
Why does the release matter if the results are still claims?
Some manuscripts address problems with large bodies of research built around them. If a result is correct and its methods are useful, it could affect future work. Review will determine whether the claims are sound and whether researchers can learn from the proofs.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
