🔍 Read the full analysis: The Promise And Uncertainty Of OpenAI’s AI Mathematics In 722 Proofs on ThorstenMeyerAI.com
Get office and shipping supplies 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, grouped into 372 families, generated by an unnamed model from about 4,000 problems. The collection includes claims about major open problems, but OpenAI says outside mathematicians have not confirmed them, and its repository warns that some results without formal verification may contain issues. The key test is whether researchers can verify and understand the work well enough to build on it.
OpenAI published 722 mathematical manuscripts on Monday, presenting results generated by an unnamed, unreleased model and spanning 372 families of related work. The collection includes claims about major open problems, but the results have not been confirmed by outside mathematicians, and the company’s repository warns that some work without formal verification may have issues.
According to OpenAI’s post and the project’s GitHub repository, the manuscripts came from a pool of roughly 4,000 problems, which the company filtered for what it described as an appropriate level of significance. OpenAI says the average result took about three hours of ChatGPT Pro thinking compute. The manuscripts cover areas including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics, and were published under the Apache-2.0 license.
Among the most striking claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, a proof that all nonabelian free group factors are isomorphic, and a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. The catalogue also includes claimed results on the Hodge conjecture for CM abelian varieties and the Mahler conjectures. These are claims in the manuscripts, not independently established solutions.
OpenAI’s materials say many, but not all, results have Lean formalizations. The repository specifically cautions that some unformalized results could have issues. The company published ten abridged reasoning summaries for the 372 families. The Riemann zero-free-region write-up was edited by humans for readability, and OpenAI identifies that and the Hodge result as 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.
Why Verification Shapes the Impact
The immediate importance of the release is not simply how many famous conjectures appear in the catalogue. It is whether mathematicians can check the arguments, identify what is new and explain the methods. A correct proof may settle a question without giving researchers tools they can use elsewhere; a proof that exposes a reusable technique could have a much wider effect.
The Unique Games Conjecture illustrates the possible reach. A substantial body of theoretical computer science studies the limits of approximation algorithms under assumptions about that conjecture. If the claimed proof is sound and matches the problem as mathematicians understand it, researchers would need to examine what follows for that work. Those consequences are conditional: the announcement alone does not establish that the conjecture has been resolved.
For readers outside mathematics, the distinction is between an answer and a discovery others can use. Researchers must be able to inspect the chain of reasoning, translate machine output into a form the field can evaluate, and determine whether it advances understanding. Independent scrutiny, rather than the size of the release, will determine its standing.
As an affiliate, we earn on qualifying purchases.
What Earlier Releases Show
This is OpenAI’s fourth major mathematics release this year, according to the source material. The earlier announcements offer examples of both possible outcomes and of why verification matters.
In May, OpenAI’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 version they described as digested and human-verified. That process made the work easier for mathematicians to assess and is one model for turning generated output into accepted mathematics.
OpenAI’s August release, titled “Ten Advances,” had a disputed result: a critique of its claimed counterexample to Connes’s rigidity conjecture said the constructed groups did not meet the condition required by the conjecture. In September, the company announced a Lean-formalized Navier–Stokes result produced using about 10,000 concurrent agents over 88 hours. The announcement prompted a dispute over research priorities; 25 Fields Medalists later signed a declaration criticizing the use of famous problems as AI benchmarks when the work does not support human understanding. Their objection concerned the purpose and practice of the research, not a finding that the Navier–Stokes proof was false.
“Digested, human-verified.”
— Five mathematicians reviewing the Erdős unit-distance result
As an affiliate, we earn on qualifying purchases.
Which Manuscripts Will Hold Up
It is not yet clear which of the 722 manuscripts are correct, which claims will survive independent review, or whether any result will be accepted as a resolution of a named open problem. OpenAI’s filtering of the roughly 4,000 prompts was conducted by the company; the source material does not describe an independent process for choosing the published results.
Formal verification can help check a proof encoded in a system such as Lean, but only some results have formalizations, and formal checking does not by itself establish that a result is important, novel or a match for the mathematical question at issue. The catalogue’s ten abridged summaries also cover only a small portion of the 372 families. How much review each manuscript receives, and how long that work will take, remains unclear.
As an affiliate, we earn on qualifying purchases.
Mathematicians Begin the Review
The next step is for mathematicians to examine individual manuscripts, check their statements and arguments, and determine whether formalizations can be completed or independently verified. Some work may be corrected, rejected or accepted after scrutiny; the release does not set a timetable for those judgments.
Researchers will also need to assess whether any valid proofs contain methods that can be understood and reused. The Erdős result provides one example of human review turning generated output into a digestible proof, while the disputed Connes claim shows why matching a proof to the exact conjecture matters. The catalogue’s significance will emerge result by result, as outside researchers report what holds up and what, if anything, the work makes possible next.
mathematical physics reference books
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, arranged in 372 families of related results. The company says they were generated by an unnamed model from roughly 4,000 problems.
Are the claimed proofs confirmed?
No. OpenAI’s CEO said the results had not yet been confirmed by outside mathematicians. The company’s repository also warns that some results without formal verification could have issues.
Which major problems do the manuscripts address?
The collection includes claims concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Riemann zeta function, the Hodge conjecture for CM abelian varieties and other topics. These are claims awaiting independent assessment.
What would make the release important beyond solving problems?
Researchers would look for methods they can understand, verify and reuse. A proof may settle a question yet have limited influence if it does not provide ideas that help advance other work.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
