Latest update
A day after Lean certified the Riemann proof, a claim that it's "invalid" hits 4,800 likes. Top reply: "not credible"
First seen on X 29 hours ago@Polymarket ♥ 4,74916 views
"Lean isn't the problem. The problem is lean is not proving unconditional RH." Jenny Lorraine Nielsen, who describes herself as a mathematician, posted that on X on Nov 21, the same day the Lean and Mathlib communities certified that "both the statement and the proof" of the Riemann Hypothesis by Anthropic mathematician Levent Alpöge, made with Claude, are correct. On Nov 22, Polymarket posted that "new analysis by mathematician Jenny Lorraine Nielsen claims the AI-assisted proof of the Riemann Hypothesis is invalid, despite its successful verification in Lean," drawing 4,800 likes and 510,000 views.
Nielsen targets one definition inside the 17-million-line codebase. She argues that the file OverIx.lean defines a geometry condition, anyGeometry, as true for any input, and that this empty condition is carried through several steps into the final hypothesis bundles, so the proof reaches its conclusion with a "true" label instead of an actual geometric object. On Nov 20, the day the proof was released, she had already written: "Alpoge's proof is also dead."
Pushback was louder. "I checked this. She's not credible. She has her own proof for GRM... She has published similar 'takedowns' of other llm generated math proofs like Navier Stokes," one reply said, drawing 810 likes. Nielsen's X bio describes her as the author of a gauge-gravity unification paper. A defender replied that she is "one of the few people explaining the mathematics behind the Lean codes."
The real question is what Lean certification guarantees. The Lean math blog Xena explained on Nov 12 that a theorem's statement and its proof must be judged separately: a mis-stated theorem still compiles, but no proof of a false statement will. For AI-written code, it said, one must first check that the statement matches the real theorem and that no extra axioms were added. The joint Lean and Mathlib statement says all four of that post's criteria are satisfied, and one participant reported that all 175,000 modules compiled using only three standard axioms. A post sharing a note on Alpöge's work by 2010 Fields medalist Ngo Bao Chau also drew 3,300 likes; the poster summarized it as saying AI proofs don't give understanding by themselves, "you still have to dig into them."
Comments 0
Select a sentence in the story (long-press on phones) to comment on it.
“BREAKING: New analysis by mathematician Jenny Lorraine Nielsen claims the AI-assisted proof of the Riemann Hypothesis is invalid, despite its successful verification in…”
