🔍 Read the full analysis: OpenAI’s AI Mathematics: One Question Looms Over 722 Proofs 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 says an unnamed, unreleased model generated 722 mathematical manuscripts across 372 families of results, selected from roughly 4,000 problems. The catalogue includes claims involving major open problems, but outside mathematicians have not yet confirmed them all; whether the work yields reusable ideas remains an open question.
OpenAI published 722 mathematical manuscripts on Monday, attributing them to an unnamed, unreleased model that worked on roughly 4,000 problems. The papers span 372 families of related results and include claims about longstanding open problems, but outside mathematicians have not yet confirmed the results; their significance will depend both on whether the proofs hold and on whether people can understand and build on them.
OpenAI’s post and GitHub repository describe manuscripts across number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. The repository makes the material available under the Apache-2.0 license. OpenAI says the work was drawn from about 4,000 problems, then filtered for what it considered an appropriate level of significance. That selection was made by the company, not by independent mathematicians.
The catalogue includes claimed work on the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Hodge conjecture for CM abelian varieties and a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. These are descriptions of claims in the manuscripts, not findings independently established by the release. The source report says that many, but not all, results have Lean formalizations, a computer-checkable representation of mathematical proof.
OpenAI’s README warns that some unformalized results could have issues. The repository also contains just ten abridged reasoning summaries for 372 result families. The source report says the write-up on the Riemann zeta function was edited by humans for readability, and identifies the Riemann and Hodge manuscripts as exceptions to the usual process. OpenAI said each result took an average of about three hours of ChatGPT Pro thinking compute; the report does not provide a complete independent accounting of the model’s work or the review behind each manuscript.
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.
Proofs Must Become Usable Ideas
For mathematicians, a correct proof can settle a question. But the lasting value of a proof often lies in the methods it introduces and the work other researchers can do with them. A catalogue of machine-generated answers would have a different impact depending on whether researchers can extract techniques, verify arguments and use them in further results.
The source report points to OpenAI’s May release on the Erdős unit-distance conjecture as a possible model: the company’s model produced a counterexample, and five mathematicians then published what they described as a digested, human-verified version. That process made the result easier for the field to assess. By contrast, a proof may be correct yet difficult to survey or reuse, or may turn out to address a statement that differs from the conjecture at issue.
One claimed result could matter well beyond its own question if it withstands review. The Unique Games Conjecture is connected to a substantial body of theoretical computer science that uses it to establish limits on approximation algorithms. If a proof were confirmed, researchers would need to examine which results relying on the conjecture are affected and how. At present, that is a conditional implication, not a verified consequence of OpenAI’s release.
As an affiliate, we earn on qualifying purchases.
A Mixed Record From Earlier Claims
This release follows three earlier OpenAI mathematics announcements described in the source report. In May, the company said its model had produced a counterexample to the Erdős unit-distance conjecture, a problem dating to 1946. Five mathematicians posted a human-verified account that day. The episode showed how researchers can turn model output into a form that can be scrutinized and discussed.
In August, OpenAI announced ten advances. One claimed counterexample, concerning Connes’s rigidity conjecture, was disputed within a day: a critique argued that the constructed groups did not meet the condition required by the conjecture. The report also says multiple independently generated counterexamples to the same conjecture were in circulation. That history makes independent checking especially relevant when a large catalogue presents many results at once.
In September, OpenAI announced a Lean-formalized proof concerning finite-time blow-up in the Navier–Stokes equations, a Millennium Prize problem. The source report says the work used about 10,000 concurrent agents over 88 hours and came amid a priority dispute involving separate work on forced Euler equations. Three days later, 25 Fields Medalists signed a declaration criticizing AI mathematics efforts that treat famous problems as benchmarks without human understanding. Their objection, as described in the source, concerned the purpose and practice of mathematical research, not a finding that the Navier–Stokes proof was false.
formal verification tools for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks Still Pending
The release does not establish that all 722 manuscripts are correct. The source report says many results have Lean formalizations, but not all, and OpenAI itself warns that unformalized work may contain problems. Formalization can help check a proof encoded in a specified system, but the catalogue’s limited summaries also leave a practical question: how readily can outside researchers understand and assess each result?
It is also unclear how OpenAI ranked the roughly 4,000 problems, what standards it applied when deciding that a result was significant, and how much human input each manuscript received. The source material does not provide independent assessments of the full set or a timeline for such reviews. Until those checks take place, claims about major conjectures should not be treated as settled mathematical results.
Even after verification, the catalogue’s broader value will remain uncertain. A correct proof may yield techniques that others can use, settle a question without changing nearby research, or fail to produce new insight. Those outcomes require time and engagement from mathematicians; publication alone cannot establish which will follow.
AI-assisted mathematical research books
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Researchers Will Test the Manuscripts
The next step is independent review: specialists can examine individual arguments, compare each manuscript’s statement with the problem it claims to solve, and test formalized portions where available. The source material does not identify a single review body or set a timetable for that work, so scrutiny is likely to proceed result by result rather than through one verdict on the entire collection.
For the most consequential claims, confirmation would require more than a striking title. Researchers would need to establish that the proof is sound, that it addresses the relevant open problem as stated, and that its reasoning can be communicated and checked. Whether any of the 372 families generates reusable methods or changes other research remains a longer-term question.
computer algebra systems for research
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 of related results and attributed to an unnamed model that has not been released.
Have mathematicians confirmed the results?
Not as a collection. The source report says the claims have not yet been confirmed by outside mathematicians, and OpenAI warns that some unformalized results could have issues.
What is the Unique Games Conjecture claim?
One manuscript claims a proof of the Unique Games Conjecture, an important open problem in theoretical computer science. The claim remains unverified, so implications for work that assumes the conjecture are conditional.
Why do researchers care whether a proof is understandable?
A proof can settle a problem, but methods that other mathematicians can understand and reuse may have greater influence on future research. Verification and useful explanation are separate parts of evaluating the catalogue’s impact.
What happens next?
Researchers are expected to scrutinize individual manuscripts, including whether their proofs are sound and match the problems they address. The source material gives no timetable for independent review.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
