🔍 Read the full analysis: 722 Proofs, One Question: Will Any Of OpenAI’s AI Mathematics Actually Lead Anywhere? 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 has published 722 mathematical manuscripts in 372 families, generated by an unnamed, unreleased model from roughly 4,000 problems. The catalogue includes claims about major open questions, but OpenAI says outside mathematicians have not confirmed them; whether the work produces useful new methods remains unknown.
OpenAI published 722 mathematical manuscripts on Monday, presenting results generated by an unnamed, unreleased model across 372 families of related work. The collection includes claims about major open problems, but OpenAI and its chief executive, Sam Altman, have stressed that outside mathematicians have not yet confirmed the results, leaving their validity and wider value unsettled.
The manuscripts span number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics, according to OpenAI’s post and repository. The company says the work came from roughly 4,000 problems posed to the model, then filtered for what OpenAI judged an appropriate level of significance. The average result used about three hours of ChatGPT Pro thinking compute. The selection process was conducted by OpenAI, not by an independent panel.
Among the claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, and a proof that all nonabelian free group factors are isomorphic. The catalogue also includes a proposed zero-free region for the Riemann zeta function to the right of Re(s) = 11/12, a result concerning the Hodge conjecture for CM abelian varieties, and work on Mahler’s conjectures. These are claims in manuscripts, not established solutions.
OpenAI released the work under an Apache-2.0 license. Lean formalizations are included for many, but not all, results. The repository README cautions that some results without formalization could have issues. The company supplied ten abridged reasoning summaries, covering only a portion of the 372 families. Its Riemann write-up was edited by humans for readability, and the Hodge result and Riemann result were exceptions to the usual process, according to the source account.
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 AI Proofs to Usable Mathematics
The practical importance of the release depends on more than whether a statement is true. Mathematicians often value a proof for the techniques and concepts it makes available, not just for resolving a question. If experts can extract a new method from an AI-generated argument and apply it elsewhere, the work could have consequences beyond the individual theorem. If a proof checks but offers no reusable insight, its effect may be narrower.
The Unique Games Conjecture illustrates the possible stakes. Theoretical computer science has many results whose limits on approximation algorithms depend on the conjecture. A verified resolution could change how researchers understand those limits and which algorithms can be considered optimal under particular assumptions. That impact is conditional: the manuscript must withstand scrutiny, and its precise conclusions must match the conjecture as mathematicians use it.
The release also raises a question about how mathematical progress is measured. A large number of claimed results may show a system’s capacity to generate arguments, but volume alone does not establish discovery. Independent checking, explanation and follow-up work will determine whether the collection changes research practice or mainly adds claims that specialists must sort through.
mathematics problem solving software
As an affiliate, we earn on qualifying purchases.
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 material. In May, its system produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians—Noga Alon, Thomas Bloom, Tim Gowers, Daniel Litt and Will Sawin—posted a digested version they described as human-verified. That example showed one route by which machine output can become legible and useful to the field: researchers translate the argument into a form they can evaluate, then check it.
An August release called “Ten Advances” had a more mixed reception. A claimed counterexample to Connes’s rigidity conjecture was challenged within a day; the critique argued that the constructed groups did not meet a condition required by the conjecture. The source account also says multiple independently generated counterexamples to the same conjecture have circulated, underlining the need to verify both the reasoning and whether it addresses the right mathematical statement.
In September, OpenAI announced a Lean-formalized argument about finite-time blow-up for the Navier–Stokes equations, generated using about 10,000 concurrent agents over 88 hours, according to the company. That work arrived amid a priority dispute involving separate research on forced Euler equations. Three days later, 25 Fields Medalists signed a declaration titled “A Severe Misalignment of AI in Mathematics.” Their stated concern, as reported in the source material, was that benchmark-driven attempts to solve famous problems without human understanding could work against mathematics’ aims. The disagreement is about the role and value of the work, not proof that the Navier–Stokes result is false.
Which Manuscripts Will Hold Up?
The central unknown is whether the manuscripts’ arguments are correct. No independent confirmation is supplied for the catalogue as a whole, and the source material does not identify outside reviewers who have completed checks across its 372 families. Lean formalization can help verify a formalized argument inside its stated framework, but many results are not formalized, and formal checking does not by itself show that a result answers the intended mathematical question or yields useful insight.
It is also unclear how OpenAI ranked the roughly 4,000 problems or how many candidates were rejected. The company selected the problems it considered significant, so the published set is not an independently chosen sample of the model’s mathematical output. Ten abridged summaries cannot convey the reasoning behind every family, and the human editing of the Riemann write-up makes it important to distinguish the model’s work from editorial changes.
Even manuscripts that prove correct may have different consequences: some could lead to new techniques, some may settle questions without generating follow-on theory, and some may fail scrutiny or address a subtly different statement. The downstream value cannot yet be inferred from the number of papers or the prominence of the problems they name.
Independent Checking Will Set the Pace
The next step is for mathematicians to examine individual manuscripts, reproduce their arguments and compare each claimed result with the exact open problem it purports to resolve. Researchers will also need to extract and explain the methods in forms that others can use. The earlier Erdős example suggests that a human-digested, independently checked proof can help bridge the gap between generated output and field-level knowledge.
OpenAI has not, in the material provided, announced a timetable for external review, named an independent verification group or specified which results will be examined first. The collection’s publication makes the work available, but it does not settle the claims. The clearest measure of progress will be public, detailed validation—and evidence that other mathematicians can build on the results.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, arranged in 372 families, generated by an unnamed model from roughly 4,000 problems, according to the company and its repository.
Have the claimed results been verified?
Not as a collection. OpenAI’s Sam Altman said the claims had not yet been confirmed by outside mathematicians. Individual results require independent scrutiny.
Does Lean formalization prove a result is useful?
A Lean formalization can help check a proof within a formal system, but it does not alone establish that a result is important, answers the intended question or offers a method other researchers can reuse. OpenAI says not all manuscripts are formalized.
Why is the Unique Games Conjecture claim attracting attention?
Many theoretical computer science results rely on the conjecture when describing limits on approximation algorithms. If a proof is verified and has the claimed scope, it could affect that body of work. That consequence remains conditional on expert review.
What would show that the release leads to meaningful discoveries?
Independent confirmation would establish whether particular claims hold. Evidence of broader impact would come from mathematicians extracting reusable ideas or techniques and applying them in further research.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
