🔍 Read the full analysis: Will OpenAI’s AI Mathematics Find Meaningful Direction In 722 Proofs? 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 says an unreleased, unnamed model produced 722 mathematical manuscripts, grouped into 372 families, from about 4,000 problems. The results include claims about major open problems, but outside mathematicians have not yet confirmed them; their lasting value will depend on verification and whether researchers can build on the methods.
OpenAI published 722 mathematical manuscripts on Monday, presenting work attributed to an unreleased, unnamed model that the company says tackled about 4,000 problems. The manuscripts make claims involving prominent open questions, but OpenAI chief executive Sam Altman has cautioned that outside mathematicians have not confirmed the results.
The manuscripts are organized into 372 families of related results and cover fields including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI says each result took an average of about three hours of ChatGPT Pro thinking compute. The collection is published under the Apache-2.0 license, and the repository includes Lean formalizations for many, but not all, of the results.
Among the 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 claims concerning the Hodge conjecture for CM abelian varieties and the Mahler conjectures in convex geometry. These are claims in the manuscripts, not results independently established by the mathematical community.
OpenAI says it selected the problems from roughly 4,000 candidates based on what it considered an appropriate level of significance. The source material says no outside group made that selection. The release includes ten abridged reasoning summaries for the 372 families. OpenAI’s repository warns that some results without formal proofs may have issues; the Riemann zero-free-region write-up was also 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.
The key question is not just whether a statement is true. In mathematics, a proof can matter because it gives researchers a technique they can reuse, connects separate areas, or makes new questions possible. A correct result that cannot be understood or extended may settle a problem without substantially changing the field.
The source material points to OpenAI’s May work on the Erdős unit-distance conjecture as a model for productive follow-up: mathematicians produced a digestible, human-verified account of the model’s counterexample. By contrast, a disputed result can consume time if researchers must first identify whether its argument proves the claim as stated. The value of this release will depend on independent checking and human understanding, not the size of the catalogue alone.
Some claims, if verified, could have consequences beyond the immediate problem. A proof of the Unique Games Conjecture, for example, could affect theoretical computer science results that rely on it when establishing limits on approximation algorithms. That potential impact remains conditional: the manuscript’s claim must first survive expert scrutiny, and researchers would then need to examine its implications.
As an affiliate, we earn on qualifying purchases.
Earlier Releases Offer a Caution
This is described in the source material as OpenAI’s fourth major mathematics release this year. The first, in May, concerned the Erdős unit-distance conjecture. Five mathematicians posted what they described as a digested, human-verified version of the model’s counterexample on the same day, illustrating how machine-generated work can be made assessable by specialists.
An August release called “Ten Advances” had a more mixed reception. A claimed counterexample to Connes’s rigidity conjecture was challenged within a day: a critique argued that the constructed groups did not meet the condition required by the conjecture. The source also reports that multiple machine-generated counterexamples to the same conjecture have circulated, reinforcing the need to check whether a proof addresses the exact mathematical statement.
In September, OpenAI announced a Lean-formalized Navier–Stokes result produced using about 10,000 concurrent agents over 88 hours. The work prompted a dispute over priority and, three days later, a declaration signed by 25 Fields Medalists criticizing the use of famous problems as benchmarks without enough attention to human understanding. The dispute concerned the purpose and practice of AI mathematics; it does not by itself establish whether the proof is correct.
formal proof verification software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Verification Across 372 Result Families
No independent confirmation of the headline claims is provided in the source material. It is not clear which of the 372 families have been reviewed by specialists, how many manuscripts have complete formalizations, or whether outside researchers can reproduce the arguments. Formal verification can check that a proof follows from encoded assumptions, but it does not by itself show that the assumptions and formalized statement match the intended conjecture.
The selection process also leaves questions about representativeness. OpenAI chose which problems from the roughly 4,000 attempts were included, and the source says only ten abridged reasoning summaries were supplied for the 372 families. The criteria for selection and the reasoning behind many individual claims are not fully described here. It remains unknown which results will prove correct, which may need revision, and which could lead to work other mathematicians can build on.
mathematical problem solving tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Expert Review Will Set the Course
The next step is for mathematicians outside OpenAI to examine individual manuscripts, check the statements and arguments, and, where possible, verify the Lean formalizations. Researchers may also try to translate promising machine-generated reasoning into proofs that are readable and useful to specialists. The source material gives no timetable for that review and does not identify a formal external review process.
As assessments emerge, the most informative measure will be the record for specific results: whether the claims hold, whether corrections are needed, and whether the methods yield further work. Until that process happens, the release is evidence of a large body of AI-generated mathematical claims—not confirmation that its most ambitious results are established discoveries.
AI-assisted theorem proving software
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, attributed to an unreleased, unnamed model. The work was drawn from about 4,000 problems, according to the source material.
Have mathematicians verified the claimed breakthroughs?
Not according to the source material. It says the claims have not yet been confirmed by outside mathematicians. Each result requires its own review.
What is a Lean formalization?
Lean is a proof assistant used to encode mathematical statements and arguments in a form that software can check against formal rules. OpenAI’s repository includes formalizations for many, but not all, of the released results.
Why does it matter whether researchers understand the proofs?
A verified answer can settle a question, but a proof’s broader value often comes from methods other researchers can understand and reuse. The release’s lasting impact will depend on whether specialists can validate and build on its arguments.
Which claim could have broad consequences if verified?
The claimed proof of the Unique Games Conjecture could affect theoretical computer science results that use the conjecture to establish limits on approximation algorithms. That impact is conditional on the proof being correct and on later analysis of its implications.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
