🔍 Read the full analysis: OpenAI’s AI Mathematics And The Question Behind 722 Proofs on ThorstenMeyerAI.com
Get the little things that make your day 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 says an unnamed, unreleased model produced 722 mathematical manuscripts across 372 families of results, selected from roughly 4,000 problems. The manuscripts include claims about several famous open problems, but OpenAI says the results have not been confirmed by outside mathematicians; whether they are correct or lead to reusable ideas remains unsettled.
OpenAI published 722 mathematical manuscripts on Monday, saying they were produced by an unnamed model that has not been released publicly. The collection spans 372 families of results selected from about 4,000 problems, including claims about major open problems; OpenAI and outside mathematicians have not established that the claims are correct.
The manuscripts cover areas including number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. OpenAI says it selected the problems for what it considered an appropriate level of significance. The source material reports that the average result used about three hours of ChatGPT Pro thinking compute. The papers are published under the Apache-2.0 license, and the repository includes Lean formalizations for many, but not all, results.
Among the claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, and results concerning free group factors, the Riemann zeta function, the Hodge conjecture for certain abelian varieties, and conjectures in convex geometry. These are claims in manuscripts, not independently validated breakthroughs. OpenAI’s repository README cautions that some results without formal proofs could have issues.
The release includes 10 abridged reasoning summaries for the 372 families. The source says two manuscripts followed exceptions to the usual process: the Riemann-related write-up was edited by humans for readability, and the Hodge result also had a different process. The details provided do not establish the extent of human involvement in every paper or explain fully how the selected results were assessed before publication.
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 Shape the Claims’ Value
The immediate test is not the number of papers or the fame of the problems named in them. It is whether specialists can check the arguments, confirm that each result matches its stated claim, and explain the reasoning. A proof that survives scrutiny can settle a question; a proof that also supplies a reusable method may influence later work. Neither outcome follows simply from publication.
The source material points to OpenAI’s May result on the Erdős unit-distance conjecture as an example of a process that can work: mathematicians produced a digested, human-verified version of the model’s output. By contrast, it describes an August claim about Connes’s rigidity conjecture that was challenged because the constructed groups reportedly did not meet a required condition. Those precedents underline why outside review matters, especially when a release contains hundreds of separate lines of argument.
There is also a broader question about what counts as progress in mathematics. The source describes a concern raised by 25 Fields Medalists after OpenAI’s September Navier–Stokes announcement: using famous problems as benchmarks without developing human understanding may conflict with the aims of mathematical research. Whether the current manuscripts offer insight, merely settle statements, or fail review will depend on what researchers can verify and learn from them.
mathematics problem solving software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Earlier Releases Offer a Mixed Record
This is described in the source as OpenAI’s fourth major mathematics release this year. In May, the company’s model produced a counterexample to the Erdős unit-distance conjecture, and five mathematicians posted a human-verified account the same day. The source presents that work as an example of machine output being translated into a form the mathematical community could evaluate.
OpenAI’s August release, called “Ten Advances,” had a more contested result: a claimed counterexample to Connes’s rigidity conjecture was challenged within a day. In September, OpenAI announced a Lean-formalized Navier–Stokes result generated using about 10,000 concurrent agents over 88 hours. The announcement also prompted a priority dispute involving separate work on forced Euler equations and a declaration by 25 Fields Medalists criticizing the use of famous problems as AI benchmarks without human understanding. These episodes provide context, but they do not verify or disprove the new manuscripts.
Formalization can help make a proof’s steps checkable by software, but the source says formal versions are available for many, not all, of the current results. It also reports that several high-profile manuscripts have Lean formalizations. That is relevant evidence about how some arguments may be checked, but it does not by itself establish that every formalized result is a meaningful resolution of the intended mathematical question.
Independent Checks Remain Outstanding
The central unknown is whether the manuscripts’ arguments are correct, complete and aimed at the precise statements mathematicians regard as open. The supplied source material reports no outside confirmation of the new collection’s high-profile claims. It also does not say how many papers have been examined by independent specialists, whether any have been challenged, or whether researchers have reproduced the results.
Other questions concern the selection and review process. OpenAI chose which problems met its significance threshold, and the source says only 10 of the 372 families have abridged reasoning summaries. The available information does not explain the selection criteria in detail, establish how much human input each manuscript received, or show whether the model’s compute time is comparable across results. Formalization may assist verification, but the repository’s own caution means readers should not infer that every claim has passed a proof check.
Mathematicians Must Test the Manuscripts
The next step is independent mathematical review: specialists will need to examine the manuscripts, check formalized arguments where available, and identify whether the proofs establish the claims as stated. Some results may be corrected, clarified or rejected; others could be verified. The source material does not provide a timetable for that process or name an external review body overseeing the collection.
For claims that hold up, a further test is whether mathematicians can extract methods and build on them. The Erdős example in the source shows how human analysis can turn model output into a digestible result. Until comparable work is done across the new collection, the number of manuscripts is a measure of what OpenAI has published, not a measure of confirmed mathematical discoveries.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, grouped into 372 families of related results and attributed to an unnamed model that has not been publicly released.
Are the claimed proofs confirmed?
No outside confirmation is reported in the supplied material. OpenAI’s stated position is that the results are claims awaiting confirmation by mathematicians, and its repository warns that some unformalized results could have issues.
What famous problems do the manuscripts address?
The reported claims include results about the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Riemann zeta function, the Hodge conjecture for certain abelian varieties, and other problems in mathematics. Their inclusion does not establish that any claim is correct.
Why does formal verification matter here?
Lean formalizations can help check that steps in an argument follow within a formal system. The source says many, but not all, results have formalizations. Formalization is useful evidence for checking; it is not, on its own, proof that a paper’s claim matches the intended open problem or yields useful new mathematics.
What happens next?
Mathematicians must independently inspect the papers and determine which arguments are sound, which need revision, and which may fail. The source does not give a review timetable or report that an external group has taken responsibility for evaluating all 372 families.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
