🔍 Read the full analysis: 722 Proofs Later, Is OpenAI’s AI Mathematics Going Somewhere? on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
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; their lasting value will depend on verification and whether people can extract reusable ideas.
OpenAI published 722 mathematical manuscripts on Monday, reporting that an unnamed, unreleased model produced them across 372 families of results after working on roughly 4,000 problems. The collection includes claims about several famous open problems, but outside mathematicians have not confirmed the results, and the release leaves open whether they will lead to new mathematical ideas or mainly require extensive checking.
OpenAI says the manuscripts span number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. The company reports that the average result used about three hours of ChatGPT Pro thinking compute. The manuscripts are published under an Apache-2.0 license, and many—but not all—include Lean formalizations, which can help check mathematical arguments in a proof assistant.
Among the collection’s most striking claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, results concerning nonabelian free group factors, and a proposed zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. The catalogue also includes claims about the Hodge conjecture for CM abelian varieties and the Mahler conjectures. These are claims made in the released manuscripts, not results independently established by the mathematical community.
OpenAI says it selected the published work from about 4,000 problems for an “appropriate level of significance.” That screening was conducted by the company; the release does not describe an independent selection process. It provides only 10 abridged reasoning summaries for the 372 families. The repository itself cautions that some results without formalizations could have issues. OpenAI also says the Riemann write-up was edited by humans for readability.
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 Useful Ideas
A correct proof can settle a mathematical question, but its influence often depends on whether other researchers can understand and reuse its methods. Verification is only one step: mathematicians must also determine whether a proof offers tools that can be applied elsewhere, or whether it establishes a result without changing how the field works.
The Unique Games Conjecture could have practical consequences for theoretical computer science if the claimed proof withstands review. A substantial body of work uses the conjecture to establish limits on approximation algorithms, including results related to the Goemans–Williamson algorithm for Max-Cut. If the claim is correct, researchers would need to examine which conclusions follow and whether existing conditional results can be strengthened or reframed. The manuscript’s existence alone does not settle those questions.
The collection therefore tests more than an AI system’s capacity to produce mathematical text. Its importance will turn on whether independent experts can validate the claims and whether they can digest the arguments into ideas other mathematicians can build on. Without that work, even a correct proof may have limited influence beyond resolving its stated problem.
proof assistant software for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Math Claims
The release follows several mathematics announcements from OpenAI this year, with mixed outcomes reported in the supplied source. In May, the company’s model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—posted a human-verified account the same day. That process, in which researchers turn machine output into an argument the field can evaluate, offers one example of how AI-generated work may become usable mathematics.
OpenAI’s August release, called “Ten Advances,” drew a dispute over a claimed counterexample to Connes’s rigidity conjecture. A critique argued that the constructed groups did not satisfy a condition required by the conjecture. In September, OpenAI announced a Lean-formalized result about finite-time blow-up in the Navier–Stokes equations. The announcement prompted debate about priority and about the role of AI in mathematical research. According to the supplied source, 25 Fields Medalists signed a declaration criticizing the use of famous problems as benchmarks when the work lacks human understanding; their objection was not presented as a finding that the Navier–Stokes proof was incorrect.
Those episodes show why the current catalogue needs careful treatment. A formalization can support checking, but it does not establish that every manuscript is correct or that a result is significant. The repository says some manuscripts are not formalized, and even a formally checked statement must be examined to confirm that it addresses the intended mathematical question.
formal verification tools for mathematical proofs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks Still Needed
No independent confirmation of the catalogue’s major claims is provided in the source material. It is unclear which manuscripts have been reviewed by outside specialists, whether any have been found correct or flawed, and how much of the collection’s technical work can be checked through Lean. OpenAI’s warning about unformalized results underscores that the degree of verification differs across the release.
The selection process also leaves questions about the catalogue’s overall picture. OpenAI chose which results to publish from roughly 4,000 problems and supplied summaries for only 10 of the 372 families. The source does not specify how many problems produced no result, how the company judged significance, or whether outside mathematicians helped select the work. It is also not yet clear whether researchers can make sense of the proofs without substantial rewriting or additional explanations.
AI-powered mathematical problem solver
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Review Will Determine the Legacy
The next step is independent mathematical review. Specialists will need to inspect the manuscripts, check that each proof supports its stated conclusion, and identify any techniques that can be reused. Lean formalizations, where available, may assist with verification, while unformalized arguments will need scrutiny through other methods.
No timetable for a comprehensive outside review is given in the source material, and OpenAI has not identified a confirmed result from the collection. The clearest early signal of lasting impact will be a manuscript that experts verify, explain in an accessible form and use to advance related work. Until then, the 722 papers are a substantial set of claims—not yet a confirmed set of mathematical discoveries.
mathematical proof verification software
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, generated by an unnamed model from roughly 4,000 problems, according to the company.
Have outside mathematicians verified the results?
The supplied source says the claims have not yet been confirmed by outside mathematicians. OpenAI’s repository also warns that some unformalized results could have issues.
What is one of the most consequential claims?
The collection includes a claimed proof of the Unique Games Conjecture, a problem with implications for theoretical computer science. The claim’s consequences depend on independent verification and further expert analysis.
Does a Lean formalization prove a result is important?
No. A Lean formalization can help check that an argument follows within a formal system, but it does not by itself show that the result is important or that its methods will be useful elsewhere.
What would show that the release has lasting value?
Evidence would include experts verifying the proofs, explaining their key ideas and applying those ideas to other problems. A correct result that offers no reusable method could settle a question without substantially changing the field.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
