🔍 Read the full analysis: OpenAI’s AI Mathematics: Asking What 722 Proofs Are Leading Toward on ThorstenMeyerAI.com
Get the latest gadgets 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 in 372 families, generated from about 4,000 problems by a model it has not named or released. The company says the results include claims about major open problems, but they have not been confirmed by outside mathematicians; the key test will be whether experts can verify and use the work.
OpenAI published 722 mathematical manuscripts on Monday, presenting work attributed to a model the company has not named or released. The collection spans 372 families of results and includes claims about major open problems, but OpenAI chief executive Sam Altman said the results have not been confirmed by outside mathematicians.
OpenAI’s post and accompanying GitHub repository describe manuscripts produced after the model was given about 4,000 problems. The company says it selected problems it considered significant. The source report says an average result required about three hours of ChatGPT Pro thinking compute. The selection and filtering were conducted by OpenAI; the material does not describe an independent process for choosing which problems made the collection.
The manuscripts cover areas including number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. Among the reported 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, Hodge theory and the Mahler conjectures. These are claims in the released material, not independently established breakthroughs.
OpenAI released Lean formalizations for many, but not all, results, according to the source report. Its repository warns that some results without formalizations could contain issues. The report also says the Riemann zero-free-region manuscript was edited by humans for readability, and that ten abridged reasoning summaries were provided across the 372 families. A proof assistant can help check formalized mathematics, but it does not by itself show that every claim is correct, that a statement matches the intended conjecture, or that the work yields useful new ideas.
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 Comes Before Discovery
The scale and ambition of the catalogue make it relevant to mathematicians and to anyone tracking the limits of AI research. But the number of manuscripts is not a measure of how many important theorems have been established. Each claim must be examined, and the quality of the proofs and their mathematical value remain open questions.
There is also a wider issue than whether a proof checks. In mathematics, a proof can matter because its methods help researchers solve other problems. The source report contrasts that kind of lasting contribution with results that settle a question without producing reusable ideas. If experts can extract methods from the OpenAI work, it could contribute to further research. If they cannot, even correct results may have limited influence beyond the statements they prove.
The distinction matters for AI evaluation, too. A system that produces a correct answer to a famous problem may perform impressively as a benchmark, yet still leave mathematicians unable to understand or extend its reasoning. The catalogue’s longer-term importance will depend on independent verification and on whether researchers can turn the work into knowledge others can use.
mathematics proof assistant software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Mixed Record in Mathematics
This is described in the source material as OpenAI’s fourth major mathematics release this year. In May, the company’s model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians then posted what the source calls a digested, human-verified version, illustrating one route from machine output to a result that researchers can assess.
OpenAI’s August release, “Ten Advances,” had a more contested reception. The source report says a claimed counterexample to Connes’s rigidity conjecture was challenged within a day because the constructed groups did not meet a condition required by the conjecture. In September, OpenAI announced a Lean-formalized Navier–Stokes result produced by a large group of agents. That announcement prompted a dispute over priority and a declaration signed by 25 Fields Medalists criticizing the use of famous problems as AI benchmarks without human understanding. Those episodes do not establish whether the new manuscripts are right or wrong; they show why review and clear statements of what has been proved matter.
The source report frames the possible outcomes in three broad ways: researchers may extract and develop useful ideas; a result may be correct but have little effect on later work; or a proof may fail or establish a claim that differs from the conjecture under discussion. The current release has not yet been sorted into those categories by independent mathematical review.
formal verification tools for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Which Claims Will Withstand Review
No outside verification of the 722 manuscripts is reported in the source material. It is not yet clear which claims will survive expert scrutiny, how long review will take, or whether independent researchers can reproduce the arguments. The repository’s warning about unformalized results adds a specific caveat, while formalization of some manuscripts does not settle the status of the entire collection.
The selection process also leaves open how representative the published work is. OpenAI says it filtered roughly 4,000 problems for an appropriate level of significance, but the source report says that choice was made internally. The materials described do not establish how many problems produced no result, what criteria were used to judge success, or whether the summaries convey enough detail for readers to assess the reasoning. It is also unknown whether any of the claimed results will lead to methods or applications beyond the immediate statements.
AI-powered mathematical problem solver
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Mathematicians Test the Papers
The next step is outside mathematical review: researchers will need to read the manuscripts, check the proofs, and determine whether the formalized versions correspond to the claims being made. For results without Lean formalizations, that work will require particular attention to the written arguments and their assumptions.
For the most consequential claims, confirmation would be only part of the process. Specialists would also need to identify what techniques, if any, can be reused and whether the results change other work that depends on the open questions. OpenAI has published the manuscripts and repository, but the source material gives no timetable for independent findings or for further company disclosures. Until reviews emerge, the claims remain unconfirmed.
theoretical computer science textbooks
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, grouped into 372 families and attributed to an unnamed model. The source report says the work followed prompts involving about 4,000 problems.
Have mathematicians verified the results?
Not according to the source material. Altman described them as claims that outside mathematicians had not confirmed. The released collection should not be treated as 722 established results.
What kinds of problems do the papers address?
The reported claims span areas including number theory, geometry, topology and theoretical computer science. They include statements about the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, and the Riemann zeta function.
Does a Lean formalization prove that a result is useful?
No. Formalization can help check that a specified argument follows within a formal system, but it does not automatically show that a manuscript proves the intended claim or provides reusable mathematical ideas. Many of the released results also do not have formalizations, according to the source report.
What would make the release important in the long term?
Independent confirmation would establish whether particular claims are correct. The release’s broader impact will depend on whether mathematicians can understand and extend the methods, rather than only verify isolated results.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
