🔍 Read the full analysis: A Look At 722 Proofs And The Road Ahead For OpenAI’s AI Mathematics on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI has published 722 mathematical manuscripts across 372 families of results, generated from about 4,000 problems by an unnamed model. The results include claims about major open problems, but they have not been confirmed by outside mathematicians; formal verification covers many, not all, of the manuscripts. Whether the work produces reusable ideas or withstands scrutiny remains unknown.
OpenAI on Monday published 722 mathematical manuscripts generated by an unnamed, unreleased model, presenting claims across 372 families of results that include solutions to several prominent open problems. The collection is a substantial new test of AI-assisted mathematics, but OpenAI says the claims have not been confirmed by outside mathematicians, and its repository warns that some results without formal proofs could contain problems.
The manuscripts cover fields including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI says the work began with roughly 4,000 problems, which the company filtered for what it considered an appropriate level of significance. The selection was made by OpenAI; the source material does not describe an independent process for choosing the problems or the results presented.
Claims in the catalogue include a proof of the Unique Games Conjecture, a result concerning Hilbert’s tenth problem over the rationals, and a proof that all nonabelian free group factors are isomorphic. Other manuscripts claim 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 results on the Mahler conjectures. These are claims in AI-generated manuscripts, not findings established by the publication itself.
OpenAI says the average result took about three hours of ChatGPT Pro thinking compute. The manuscripts are available under an Apache-2.0 license. Many, but not all, have Lean formalizations, which can help check that a proof follows specified formal rules. The collection also includes just 10 abridged reasoning summaries for the 372 families. OpenAI’s README cautions that some results without formalization could have issues.
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.
From AI Proofs to Reusable Mathematics
The stakes extend beyond whether a particular conjecture is settled. In mathematics, a proof can matter because its methods help researchers solve other problems. The collection’s long-term value will depend on whether mathematicians can verify the results, understand how they work and extract techniques that others can use.
A confirmed proof of the Unique Games Conjecture, for example, could affect theoretical computer science because many results about the limits of approximation algorithms are established under that conjecture. But the source material does not establish that the manuscript is correct, and it offers no independent assessment of the consequences if it is. The distinction between a striking claim and a dependable result is central to judging the release.
There is also a question of usability. Some work may be correct yet difficult to interpret or build on; some may prove a statement that does not match the intended conjecture; and some may fail under review. AI-generated output becomes a mathematical contribution only through the same demanding steps of checking, clarification and follow-up that apply to other work.
AI mathematical proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Earlier Releases Set the Test
The publication follows three other major OpenAI mathematics releases this year, according to the source material. In May, a model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—then published a digested version they had verified. That episode offers one possible path from machine output to a result the mathematical community can evaluate.
OpenAI’s August release, called “Ten Advances,” had a more contested outcome. A claimed counterexample to Connes’s rigidity conjecture was challenged within a day; the critique said the constructed groups did not meet a condition required by the conjecture. The source material says related machine-generated counterexamples from other groups were also in circulation. 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. That announcement prompted debate over priority and the purpose of using famous problems as AI benchmarks.
More broadly, the history of computer-assisted proof shows why correctness and mathematical influence are separate questions. The Four Colour Theorem was proved with computer assistance in 1976, but its case-checking approach has often been contrasted with proofs whose methods generate further theory. Whether any of the new manuscripts lead to reusable methods cannot be judged from their number or the prominence of the claims alone.
“A Severe Misalignment of AI in Mathematics.”
— The 25 Fields Medalists who signed “A Severe Misalignment of AI in Mathematics”
Independent Checks Still Needed
The key unknown is how many of the 722 manuscripts will survive expert review. The supplied information gives no independent assessment of the full collection, and it does not identify which claims have been checked by external mathematicians. Lean formalization can verify that a formal proof follows its encoded definitions and rules, but it does not by itself show that the formal statement matches the intended mathematical claim or that the result will be useful to researchers.
It is also unclear how much detail outside reviewers can access beyond the manuscripts and the limited set of 10 abridged summaries. The source material does not report a timetable for reviews, confirmations, corrections or withdrawals. Until those steps occur, the headline claims—including the proposed result about the Unique Games Conjecture—should be treated as unverified.
Review Will Determine the Impact
The next step is for mathematicians to examine the manuscripts, check the arguments and compare each result with the precise statement it claims to prove. Formalizations may support that work where they are available; unformalized results will need scrutiny through other methods. OpenAI’s publication of the files makes the claims available for examination, but the supplied material does not set a schedule for that review.
For readers, the most useful milestones will be independent verification, clear explanations of the methods, and follow-on work showing whether those methods apply elsewhere. A confirmed result could settle an important question; a digestible proof might also produce new techniques. If a proof fails or proves a narrower statement, that too would clarify what the model accomplished. For now, the catalogue is evidence of a large-scale mathematical effort, not confirmation that its headline claims are correct.
Key Questions
What did OpenAI publish?
OpenAI published 722 AI-generated mathematical manuscripts, grouped into 372 families of related results and made available under an Apache-2.0 license.
Are the claimed proofs confirmed?
Not on the information provided. OpenAI says the claims have not been confirmed by outside mathematicians. Many manuscripts have Lean formalizations, but not all do, and the repository warns that some unformalized results could have issues.
What are some of the major claims?
The manuscripts claim results involving the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, nonabelian free group factors, the Riemann zeta function, and the Hodge conjecture for CM abelian varieties. These remain claims pending independent review.
Why does independent review matter if some proofs are formalized?
Formal proof checking can test whether a proof follows a formal system’s rules. Mathematicians must still assess whether the formal statement captures the intended problem, whether the work is sound in context and whether its methods can be understood or reused.
What happens next?
Researchers will need to examine and verify the manuscripts. The supplied material does not give a review timetable, so the status of individual claims will become clearer as independent assessments are published.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
