Riemann proof built with Claude spans 17M lines of Lean — no human will ever read it
r0ck3t23 · x · 2026-11-23
A long-form essay argues AI has ended mathematics' status as the one field where you never had to take anyone's word for it:
- Historical anchors: Ramanujan's unsolicited theorems won over Hardy in 1913; in 1993 referee Nick Katz's repeated "I do not understand" exposed a hole in Wiles's Fermat proof. That sentence was mathematics' core safeguard.
- Precedents: the 1976 computer-checked four color theorem was rejected by many; after referees gave up on the Kepler conjecture, machine verification in 2014 made their opinions, in Hales's words, no longer matter.
- The new event: on Nov 20, Levent Alpöge released a Riemann Hypothesis proof built with Claude — 17 million lines of Lean across 175,187 files (reading 1,000 lines/day would take 47 years). Lean/Mathlib communities declared "full confidence" within a day, and independent checkers at MIT and Waterloo accepted every step. Alpöge admits limited mastery of parts of his own argument.
- Takeaway: machines now check cases, check proofs, and help write them; humans went from author to referee to audience. Echoing Musk's "unequivocal superintelligence," math now works like the rest of life — you take the machine's word for it.
More from AGI Musings
- Viral slide: "You need artists to make art. Because machines don't suffer." — ordax · 2026-11-23
- Gary Marcus Resurfaces His Three o3 Predictions as Followers Call Them Prescient — GaryMarcus · 2026-11-23
- MIT Professor: Models Cross a Discovery Threshold Around 0.76 Log-Perplexity — ProfBuehlerMIT · 2026-11-23
- Skeptic Questions OpenAI's 'Long-Term Research' Explanation for the 2026 Loss Cliff — CFGeek · 2026-11-23
- Noam Brown's 2023 bet lands: 10K agents solve Navier–Stokes, Riemann proof claimed — haider1 · 2026-11-23
- 2004 Letter Predicted Mathematicians Would Be the First Profession AI Makes Obsolete — ericwdolan · 2026-11-23