🔍 Read the full analysis: Why OpenAI’s 722 Proofs Leave The Future Of AI Mathematics Unclear 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 published 722 mathematical manuscripts across 372 families, selected from roughly 4,000 problems given to an unnamed, unreleased model. The results include claims about major open problems, but most have not been confirmed by outside mathematicians, and it is unclear whether they will produce reusable ideas or verified solutions.
OpenAI published 722 mathematical manuscripts on Monday, presenting results generated by an unnamed model that the company has not released. The papers cover 372 families of related results, including claims about major open problems; OpenAI says they are not yet confirmed by outside mathematicians.
The manuscripts span areas including number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. OpenAI’s repository says the work came from roughly 4,000 problems posed to the model and then filtered by the company for what it considered an appropriate level of significance. The average result used about three hours of ChatGPT Pro thinking compute, according to the source account.
Among the manuscripts are claimed proofs concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, and whether all nonabelian free group factors are isomorphic. Other papers claim a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12 and results related to the Hodge and Mahler conjectures. These are claims in the released work, not independently established resolutions.
OpenAI published the collection under the Apache-2.0 license. The repository includes Lean formalizations for many, but not all, results. Its README cautions that “some of the unformalized results could have issues.” The release also contains ten abridged reasoning summaries for 372 families, while the selection and significance screening were carried out by OpenAI rather than independent reviewers.
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 Their Value
The collection’s importance depends on more than whether individual claims survive checking. In mathematics, a proof can settle a question yet offer little that other researchers can use; a proof that reveals a reusable technique can influence a field well beyond the original problem. External verification and human understanding will help determine which outcome applies here.
The Unique Games Conjecture is one example of why the claims draw attention. Many results in theoretical computer science are established under assumptions connected to the conjecture, including limits on the performance of approximation algorithms. If a proof is correct and accepted, researchers would need to examine which conclusions follow and whether existing results change. The release itself does not establish that any such consequences should be revised.
The distinction also matters for how AI systems are evaluated. A machine-produced argument might be correct but difficult to interpret, or it might fail to prove the precise statement researchers care about. A checked proof is not automatically a useful discovery: researchers still need to understand its methods, identify implications and test whether those methods generalize.
AI mathematical proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Math Releases
This is described in the source account as OpenAI’s fourth major mathematics release this year. In May, a model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians then posted a human-verified account of the argument, turning machine output into a form the mathematical community could evaluate. That process is presented as one possible route from AI output to accepted research.
An August release called “Ten Advances” had a more mixed reception. A claimed counterexample to Connes’s rigidity conjecture was challenged within a day: a critique said the constructed groups did not meet a condition required by the conjecture. The episode illustrates why apparently substantial results need close review, including checks that the proof addresses the intended question.
In September, OpenAI announced a Lean-formalized result concerning finite-time blow-up in the Navier–Stokes equations, produced using about 10,000 concurrent agents over 88 hours, according to the source material. That release drew a public dispute over research priorities: 25 Fields Medalists signed a declaration criticizing the emphasis on solving famous problems as benchmarks without sufficient human understanding. Their criticism concerned the aims and practices of the work; the source does not describe it as a finding that the proof was incorrect.
“A Severe Misalignment of AI in Mathematics.”
— The Fields Medalists’ declaration, titled “A Severe Misalignment of AI in Mathematics”
As an affiliate, we earn on qualifying purchases.
Which Claims Will Survive Review
No outside confirmation is reported for the collection’s headline claims, and the source material does not give a review status for each of the 372 result families. Formalization can help check whether a proof follows from stated assumptions in a formal system, but many results in this collection are not formalized; the repository itself warns that some may have issues.
It is also unclear how the 4,000 problems were selected, what criteria OpenAI used to judge significance, and how many manuscripts have been examined by independent specialists. The source account identifies two departures from the usual process: the Riemann zero-free-region write-up was edited by humans for readability, and the Hodge result received special treatment. Those details do not establish the correctness of either claim.
Even if results are verified, their longer-term influence is unknown. Researchers would need to determine whether the proofs contain ideas that can be reused, whether they settle the stated problems in the form the field recognizes, and whether follow-on work changes existing theory.
mathematical research software for AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks and Follow-On Work
The next step is for mathematicians to inspect individual manuscripts, check formalized arguments where available, and produce clear accounts of what each proof establishes. The May Erdős result offers one example of this process: researchers translated the model’s output into a digested, human-verified version. The source material does not identify a timetable for comparable reviews of the new collection.
Researchers will also need to separate verification from impact. A manuscript may be correct while contributing little reusable theory, or its central argument may expose techniques that prompt new work. Further publications and independent assessments will show which results, if any, move from AI-generated claims to accepted mathematical contributions.
AI research notebooks for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI release?
OpenAI published 722 mathematical manuscripts, grouped into 372 families. The work was generated by an unnamed model that the company has not released, according to the source material.
Has OpenAI proved the major conjectures named in the collection?
The manuscripts contain claims about major open problems, including the Unique Games Conjecture. The claims have not been confirmed by outside mathematicians in the source account, so they should not be treated as accepted resolutions.
How were the results checked?
Many, but not all, results have Lean formalizations, which can help check formal arguments. OpenAI’s repository warns that some unformalized results could have issues. The source does not provide independent review outcomes for the full collection.
Why does it matter whether the proofs are understandable?
Mathematical proofs can be valuable not only for settling questions but also for introducing methods other researchers can use. Even a correct result may have limited impact if its reasoning is hard to interpret or does not lead to reusable ideas.
Source: ThorstenMeyerAI.com
NFL season / tailgating Picks
team gear
As an affiliate, we earn on qualifying purchases.
