🔍 Read the full analysis: OpenAI’s AI Mathematics: 722 Proofs And The Search For A Way Forward 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 produced 722 mathematical manuscripts across 372 families of results, selected from about 4,000 problems. The manuscripts include claims involving major open problems, but outside mathematicians have not confirmed them; OpenAI warns some unformalized results may have issues. The central test is whether researchers can verify the work and extract methods others can use.
OpenAI published 722 mathematical manuscripts on Monday, saying they were produced by an unnamed model it has not released. The collection spans 372 families of related results and includes claims about longstanding open problems, but OpenAI chief executive Sam Altman said the claims have not yet been confirmed by outside mathematicians.
OpenAI’s post and GitHub repository describe work across number theory, geometry, topology, operator algebras, theoretical computer science and mathematical physics. The manuscripts were selected from roughly 4,000 problems posed to the model. OpenAI says it filtered those problems for an “appropriate level of significance”; the selection was made by the company, not by independent mathematicians. The average result used about three hours of ChatGPT Pro thinking compute, according to the source material.
The catalogue includes claimed results concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12, the Hodge conjecture for CM abelian varieties and the Mahler conjectures. These are claims in the released manuscripts, not established solutions. Many results have Lean formalizations, but not all; OpenAI’s repository cautions that “some of the unformalized results could have issues.”
The release includes only 10 abridged reasoning summaries for the 372 families. The source says two manuscripts followed exceptions to the standard process: the Riemann zero-free-region write-up was edited by humans for readability, and the Hodge result was also treated differently. The materials do not, on their own, establish how much of the reasoning behind each result is accessible to researchers evaluating it.
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 Could Shape What Comes Next
The immediate question is not simply whether a model can produce a large number of mathematical claims. It is whether specialists can check the statements and proofs, identify useful techniques and explain the work in a form the field can build on. A correct proof may settle a problem; a proof whose ideas can be reused may also change how other problems are approached.
The Unique Games Conjecture illustrates the potential consequences. The source material says a substantial body of theoretical computer science relies on the conjecture to establish limits on approximation algorithms, including results concerning the classic Goemans–Williamson algorithm for Max-Cut. If the manuscript proves the conjecture as stated and withstands scrutiny, it could affect that literature. But until specialists verify the proof and its exact conclusion, those implications remain conditional.
OpenAI’s earlier releases suggest why evaluation matters. In May, researchers produced a human-digested and verified version of the model’s counterexample to the Erdős unit-distance conjecture. In August, a claimed counterexample to Connes’s rigidity conjecture was challenged because critics said the constructed groups did not meet the conjecture’s required condition. Those examples point to distinct outcomes for the new collection: work can be verified and made useful, prove correct but yield little reusable theory, or fail to establish the intended claim.
As an affiliate, we earn on qualifying purchases.
A Year of High-Profile Math Claims
The 722-manuscript release is described in the source material as OpenAI’s fourth major mathematics release this year. The May Erdős result was followed by an August package called “Ten Advances,” in which a proposed counterexample to Connes’s rigidity conjecture drew a rapid critique. The disagreement underscored that a result’s wording and hypotheses matter: a construction is not a counterexample if it does not satisfy the conditions at issue.
In September, OpenAI announced a Lean-formalized result claiming finite-time blow-up for the Navier–Stokes equations, produced using about 10,000 concurrent agents over 88 hours, according to the supplied account. That announcement also prompted a dispute over priority with concurrent work on forced Euler equations by Levent Alpöge and Tristan Buckmaster. Three days later, 25 Fields Medalists signed a declaration titled “A Severe Misalignment of AI in Mathematics.” Their concern, as described in the source, was that treating famous problems as benchmarks without developing human understanding could conflict with the aims of mathematics.
Formal verification can help establish that a proof follows from stated assumptions within a formal system, but it does not by itself show that the result is important, that the formal statement matches the intended conjecture, or that the proof offers ideas mathematicians can reuse. The source contrasts the human understanding and lasting techniques associated with proofs by Andrew Wiles and Grigori Perelman with the Four Colour Theorem’s computer-assisted proof, which settled a question through extensive case checking but is presented as a less generative model.
As an affiliate, we earn on qualifying purchases.
Independent Checks Still Needed
No outside confirmation is reported for the 722 manuscripts as a collection, and the supplied material does not identify which individual results have since been reviewed by independent experts. It is also unclear how much of the reasoning is available beyond the manuscripts and the 10 abridged summaries, or how OpenAI ranked the roughly 4,000 problems before selecting the published set.
Lean formalizations are available for many, but not all, results. For each claim, readers still need to know whether the formal statement matches the conjecture researchers care about, whether the proof is complete and correct, and whether the result is genuinely new. The source does not establish that every manuscript has been independently checked or that the headline claims will survive scrutiny.
It is also too early to tell which results, if any, will yield reusable methods. A proof can be correct without producing a technique that changes related work. That distinction cannot be settled by counting manuscripts or by the scale of the compute used.
As an affiliate, we earn on qualifying purchases.
Mathematicians Must Test the Claims
The next step is independent, result-by-result scrutiny: specialists will need to compare each manuscript with the relevant open problem, examine its reasoning, and, where available, check its formalization. For claims that withstand that process, researchers can then produce clearer accounts of the arguments and assess whether the methods can be extended to other questions.
The source material does not give a timetable for those reviews or describe a formal external evaluation process. Nor does it say when OpenAI will release the unnamed model or provide additional reasoning materials. Until those details emerge, the collection is best understood as a large set of research claims awaiting assessment—not as 722 confirmed discoveries.
theoretical computer science 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, grouped into 372 families and attributed to an unnamed, unreleased model. The company says they were selected from roughly 4,000 problems posed to the system.
Have mathematicians confirmed the claimed breakthroughs?
Not as a collection. The source says the claims have not yet been confirmed by outside mathematicians. Each result needs its own expert review, and OpenAI warns that some unformalized work could have issues.
Do the manuscripts prove the Unique Games Conjecture?
The released collection includes a manuscript claiming a proof of the Unique Games Conjecture. The source does not report independent confirmation, so it should not yet be described as a settled proof.
Why does it matter whether the proofs are understandable?
Mathematical progress often depends on methods that can be reused, not only on settling a statement. Researchers need to verify the work and determine whether its reasoning can support further results.
What happens next?
Mathematicians must examine the manuscripts, check their statements and proofs, and assess any formalizations. OpenAI has not provided a timetable for independent reviews or a release date for the model behind the work.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
