🔍 Read the full analysis: Will OpenAI’s AI Mathematics Go Further Than 722 Proofs? on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI published 722 mathematical manuscripts attributed to an unnamed, unreleased model, selected from roughly 4,000 problems. The catalogue includes claims about major open problems, but the results have not been independently confirmed as a body of work; their value will depend on expert checking and whether mathematicians can extract useful ideas.
OpenAI published 722 mathematical manuscripts on Monday, presenting work attributed to an unnamed, unreleased model and selected from about 4,000 problems. The collection includes claims about prominent open problems, but OpenAI chief executive Sam Altman said the results have not been confirmed by outside mathematicians, leaving verification and the work’s lasting value unresolved.
The manuscripts are arranged into 372 families of related results across areas including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI’s repository says many results have Lean formalizations, a computer-checkable representation of mathematical proofs, but not all do. Its README warns that some unformalized results could have issues. The work is published under an Apache-2.0 license.
The catalogue includes claims concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the isomorphism of nonabelian free group factors, and a zero-free region for the Riemann zeta function to the right of Re(s) = 11/12. It also includes work on the Hodge conjecture for CM abelian varieties and the Mahler conjectures in convex geometry. These are claims in manuscripts, not independently established solutions.
OpenAI says the model was given roughly 4,000 problems, then the company selected work it considered significant. The average result used about three hours of ChatGPT Pro thinking compute, according to the source material. OpenAI released ten abridged reasoning summaries, rather than summaries for all 372 families. The Riemann-region write-up was edited by humans for readability, and the Hodge result followed a different process from the standard workflow.
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 Will Shape the Impact
The central issue is not the number of manuscripts but whether their proofs withstand independent scrutiny and whether their methods help mathematicians make further progress. A correct result can settle a question without providing a reusable technique. The collection’s claims range from established research areas to famous, difficult open problems, so each result needs careful checking in its own mathematical context.
The difference matters for fields that rely on a conjecture as a foundation for other work. The Unique Games Conjecture, for example, is used in theoretical computer science to establish limits on approximation algorithms. If a proof of it were verified, it could affect results built on that assumption. But no such consequence follows from the manuscript’s existence alone: experts would first need to confirm that the proof is sound and addresses the conjecture as stated.
There is also a broader question about what counts as progress. A machine-generated proof may answer a question yet remain difficult for people to understand or use. The source account describes the hoped-for outcome as mathematicians digesting the work, identifying its underlying ideas and building on them. That step, not publication volume by itself, will determine whether the release changes mathematical practice.
mathematics problem solving software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Earlier Releases Offer Caution
This is described in the source material as OpenAI’s fourth major mathematics release this year. Earlier examples show why claims need to be separated from verified outcomes. In May, an OpenAI model produced a counterexample to the Erdős unit-distance conjecture; five mathematicians then published what they called a digested, human-verified version. That process offered a route for turning machine output into work the field could evaluate.
OpenAI’s August release, called “Ten Advances,” had a more contested result: its claimed counterexample to Connes’s rigidity conjecture was challenged within a day. The critique said the constructed groups did not meet a condition required by the conjecture. The source account also says several independent AI-generated counterexamples to the same conjecture have circulated, underscoring that a plausible-looking result can target the wrong formulation or contain a flaw.
In September, OpenAI announced a Lean-formalized result concerning finite-time blow-up in the Navier–Stokes equations. That release prompted a separate dispute about research priority and a declaration signed by 25 Fields Medalists, including Terence Tao, Peter Scholze and Maryna Viazovska. The signatories criticized using famous problems as AI benchmarks without human understanding; the source material says their objection was not that the proof had been shown to be wrong. Together, these episodes make verification and comprehension distinct questions.
As an affiliate, we earn on qualifying purchases.
The Manuscripts Await Expert Review
It is not yet clear how many of the 722 manuscripts will survive independent review, how quickly specialists can evaluate them, or whether the results are correct in their stated form. Lean formalization can help check a proof encoded in the system, but the source says formalizations exist for many, not all, results; that does not establish that every claim in the collection has been formally verified.
OpenAI has not named the model, and the selection process was conducted by the company rather than by outside mathematicians. The release includes reasoning summaries for only ten of the 372 families. The source material also identifies exceptions to the usual procedure for the Riemann and Hodge manuscripts, but does not provide enough detail to assess how those exceptions affected the work. No independent verdict on the collection as a whole is supplied.
It also remains uncertain whether any correct results will produce reusable methods. A proof can be valid but hard to interpret, or it can rely on an approach that does not transfer to other problems. The results’ future influence will depend on mathematical review and on whether researchers can understand, explain and extend them.
AI-powered mathematical proof assistants
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checking Comes Next
The next step is for mathematicians with relevant expertise to examine individual manuscripts, check their assumptions and verify that each proof establishes the stated result. For work with Lean formalizations, reviewers can inspect the encoded proof; other manuscripts will require scrutiny of their arguments and any supporting materials. The source material gives no announced timetable for a comprehensive external review.
As reviews appear, the most useful updates will distinguish results that are formally or independently verified from those that remain claims, are challenged, or need revision. Researchers will also need to explain whether a verified result offers a technique others can use. Until those evaluations are available, the publication is evidence that OpenAI has released a large body of mathematical work—not evidence that all, or any particular number, of its headline claims are correct.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, grouped into 372 families and selected from roughly 4,000 problems posed to an unnamed model.
Has an outside mathematician verified the results?
The source material says the claims have not yet been confirmed by outside mathematicians. The release should not be treated as a set of established solutions.
What major problems do the manuscripts address?
The catalogue includes claims concerning the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Riemann zeta function, and the Hodge conjecture for CM abelian varieties, among other topics.
Does a Lean formalization prove a result is correct?
A Lean formalization can allow a computer to check an encoded proof, but the source says many, not all, results have formalizations. It also cautions that some unformalized results may have issues.
When will the results’ impact be known?
There is no review timetable in the source material. Their impact will become clearer as specialists check the manuscripts and determine whether any verified methods can support further work.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
