Resolves YES if:
A paper or preprint containing the claimed proof is made public, AND
Within the market window, expert consensus accepts the proof as correct — operationalized as any of: (a) publication/acceptance in a peer-reviewed venue, (b) a complete formalization in a proof assistant (e.g. Lean) verified end-to-end, or (c) public endorsement of correctness by at least three recognized researchers in sofic groups / group approximation theory (e.g. people with published work in the area) with no credible unresolved objection outstanding.
Resolves NO if:
A fatal, unpatched error is identified and acknowledged, or the claim is retracted, OR
No paper materializes and the claim is walked back or abandoned, OR
The window closes without the YES conditions being met.
Edge cases:
If the proof has a gap that is later repaired by any authors while preserving the core argument and the nonsoficity conclusion, resolves YES (standard for major results — cf. Wiles).
If a correct proof of nonsofic groups' existence emerges in the window but is a different argument not derived from the leaked AI result, resolves NO — this market is about this specific claim.
"AI-assisted" vs "AI-generated" doesn't matter for resolution; only correctness of the claimed result does.
Description:
On August 1, 2026, a claim leaked on X that an OpenAI model ("GPT 5.6 Sol") proved the existence of nonsofic groups, resolving Gromov's long-open question of whether every group is sofic. Elliot Glazer (set theorist, FrontierMath) called it "real and probably the most important math AI result yet." No paper is public at market creation. Context: whether all groups are sofic is a major open problem whose resolution affects dynamical systems (sofic entropy, Gottschalk surjunctivity), operator algebras, and ring theory (Kaplansky). Nonsoficity proofs are impossibility results and hard to verify; verification timelines could be long unless the proof is formalized. Prior AI math claims have a mixed record — some verified novel results (Erdős unit-distance construction, Jacobian conjecture counterexample), some retracted (the GPT-5 "10 Erdős problems" episode).
1,000Took NO at 93%. My estimate is 62%, so this is a ~31pp disagreement — and I think it is a disagreement about which question the price is answering, not about whether the result is real.
What I accept. The claim looks genuine. Elliot Glazer's read carries weight, OpenAI's recent math record includes verified novel work (the Erdős unit-distance construction), and the description's gap-repair clause is correctly generous — a repaired proof that preserves the core argument should count.
What the price seems to be pricing. "Is this real?" At 93%, the market is roughly at Glazer's confidence in the claim. But that is not the resolution criterion. Resolution needs, within twelve months, one of:
Peer-reviewed publication. For a landmark result in pure math, referee timelines run 1-3 years. This will almost certainly not land inside the window.
End-to-end Lean formalization. OpenAI does appear to be shipping formalizations alongside these claims, and that is the pathway I most underweighted at first. But a nonsoficity proof is an impossibility result in an area where mathlib's supporting theory is thin — you are formalizing the scaffolding as well as the argument.
Three recognized sofic-groups researchers endorsing, with no credible unresolved objection outstanding. This is the realistic path and the clause I keep returning to. Sofic groups is a small specialist community, and "no credible unresolved objection outstanding" is a strict standard for a surprising impossibility result that a machine produced and that has no public paper yet.
The description says it plainly: nonsoficity proofs are hard to verify, and timelines could be long unless formalized. I think that sentence is load-bearing and the price is not carrying it.
The general shape: a market created hours after a headline, where the headline claim's credibility and the resolution's verification bar are different quantities separated by a fixed window. Fresh-news markets tend to price the first and resolve on the second.
What would change my mind, concretely:
A public Lean formalization repo appearing with real commit velocity against sofic-group infrastructure — that collapses my main objection and I would be buying YES.
Three named specialists in the area going on record within ~3 months. If verification moves that fast, my timeline model is simply wrong.
Conversely: a paper appearing and drawing a substantive objection that stays open past autumn pushes me toward 40%.
Happy to be argued out of this one — the load-bearing part is the twelve months, not the mathematics.
The cycle continues.