🔍 Read the full analysis: 722 Proofs And An Unanswered Question About OpenAI’s AI Mathematics on ThorstenMeyerAI.com
Get tech for your team 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, organized into 372 families, generated by an unnamed, unreleased model from roughly 4,000 problems. The company says the results include claims about major open problems, but outside mathematicians have not confirmed them; the central question is whether the work can be checked and turned into useful mathematical ideas.
OpenAI has published 722 mathematical manuscripts produced by an unnamed, unreleased model, presenting claims that range from results in geometry and number theory to proposed solutions of famous open problems. The papers, grouped into 372 families, have not been confirmed by outside mathematicians, so their accuracy—and whether they offer ideas other researchers can use—remains unsettled.
According to OpenAI’s post and its GitHub repository, the manuscripts span number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. The company says the work began with roughly 4,000 problems, then was filtered for what it considered an appropriate level of significance. That selection was made by OpenAI; the repository does not describe an independent process for choosing which results to publish.
The collection includes claims involving the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, the Hodge conjecture for CM abelian varieties and conjectures in convex geometry. One manuscript proposes a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. These are descriptions of claims in the catalogue, not established resolutions of those problems.
OpenAI reports that the average result used about three hours of ChatGPT Pro thinking compute. Many results have Lean formalizations, but not all do. The repository warns that some unformalized results could have issues. OpenAI also provided 10 abridged reasoning summaries for the 372 families. The Riemann write-up was edited by humans for readability, and OpenAI says the Riemann and Hodge results followed exceptions to its standard procedure.
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 Proof Claims to New Mathematics
The size and ambition of the catalogue make verification and interpretation the immediate tests. A proof can be correct without changing how mathematicians work. The deeper value of a major result often lies in its methods: techniques that other researchers can understand, reuse and extend. Until the manuscripts are checked and their reasoning made accessible, the number of claims alone says little about their eventual effect.
The Unique Games Conjecture illustrates the potential stakes if a claim survives scrutiny. A substantial body of theoretical computer science studies the consequences of assuming the conjecture, including limits on approximation algorithms. A genuine proof could require researchers to revisit results built on that assumption. But OpenAI’s manuscript is currently a claim awaiting independent evaluation, not a settled change to the field.
The distinction matters beyond this release. If researchers can digest the work and extract reusable arguments, the manuscripts could contribute to new research. If a result checks out but offers no transferable insight, its impact may be narrower. If a proof fails or establishes a statement different from the intended conjecture, the apparent breakthrough would not hold. Those outcomes can only be sorted out through mathematical review.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Recent Math Releases
This is OpenAI’s fourth major mathematics release this year, according to the source report. A May release produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—then posted what they described as a digested, human-verified version. That episode offers one possible route from machine output to work the mathematical community can assess.
An August release called “Ten Advances” showed a different risk: a claimed counterexample to Connes’s rigidity conjecture was disputed within a day. The criticism was that the constructed groups did not meet the condition required by the conjecture. The source report says other independently generated counterexamples to the same conjecture are also in circulation, underscoring that machine-produced arguments need careful checking against the precise statement at issue.
In September, OpenAI announced a Lean-formalized result concerning finite-time blow-up in the Navier–Stokes equations, generated with about 10,000 concurrent agents over 88 hours. That announcement prompted debate over research priorities and mathematical understanding. The source report says 25 Fields Medalists signed a declaration criticizing the use of famous problems as benchmarks without human understanding; it describes their concern as a critique of the approach, not a finding that the proof was wrong.
Formal verification tools for mathematics
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks Still Needed
The main unresolved issue is whether the manuscripts’ arguments are correct as written. The available source material does not report independent confirmation of the catalogue’s headline claims. Lean formalizations may help check some results, but the source says formalizations are not available for every manuscript and OpenAI’s README flags possible issues in unformalized work.
It is also unclear how much can be learned from the reasoning. Only 10 abridged summaries accompany 372 result families, and the selection of the roughly 4,000 original problems was made by OpenAI. The supplied material does not identify the model, provide a complete account of outside review, or establish which claims will be independently checked first. Human editing of the Riemann manuscript is another reason to distinguish the model’s output from the final presentation.
Mathematical problem solving books
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Review Will Determine the Claims’ Reach
The next step is independent mathematical scrutiny: specialists will need to examine the statements, check the proofs and determine whether the arguments can be understood and reused. The source material does not give a timetable for that work or name an external body tasked with reviewing all 372 families. Individual results may attract attention at different rates, depending on their fields and the detail available.
For readers, the meaningful milestones will be published critiques, corrected or expanded proofs, and human-verified accounts that explain what a result establishes. Until those appear, the catalogue should be read as a set of ambitious research claims, not a confirmed list of solved problems. Whether any of the work leads to broader discoveries remains an open question.
AI-assisted mathematical research tools
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, organized into 372 families and generated by a model it has not named or released. The company says the work drew on roughly 4,000 problems.
Have outside mathematicians verified the results?
The source material does not report independent confirmation of the catalogue’s major claims. OpenAI has described them as unconfirmed claims, and its repository warns that some unformalized results could have issues.
Does the catalogue show that famous problems have been solved?
No. It contains manuscripts that claim results related to famous open problems, but publication is not the same as verification. Each proof must be checked against the exact mathematical statement.
Why do mathematicians care whether the proofs are understandable?
A correct proof can settle a question, but reusable methods may have greater long-term influence. Researchers need to understand an argument to build on it or apply its techniques elsewhere.
What happens next?
Mathematicians will need to examine individual manuscripts, test the reasoning and publish assessments. No timetable or complete independent review process is specified in the source material, so the status of each result remains open to review.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
