NPR probe: AI math proofs passed Lean verification but proved weaker claims
burny_tech · x · 2026-10-22
An NPR investigation highlights a growing problem in AI-generated mathematics. Cambridge mathematician Anders Hansen found a discrepancy between an AI-generated proof and its Lean formalization: a constant silently changed from 4 to 5, making the formally verified statement weaker than the original claim. The Lean verification itself wasn't necessarily wrong — it may simply have verified something different from what the argument asserted. The report also features Daniel Litt, who found an error invalidating a separate AI-generated paper, and Dana Moshkovitz on a proposed solution to the Unique Games Conjecture. Key takeaway: verifying a formal proof isn't enough unless the formal statement faithfully represents the intended theorem.
More from Models
- Similarweb: Grok Drew 40.91% New-User Traffic Share in September, Highest Among Major AI Tools — XFreeze · 2026-10-22
- Rumor: 'Gemini 4 Carbon' updated checkpoint spotted in cryptic Carbon vs Argon teaser — lyraxana · 2026-10-22
- Recursive Bets on RSI: Its Eureka Machine Built the Flint Coding Model in Months — mikeflache · 2026-10-22
- 0.9B Model Transcribes a 4-Person Meeting on a MacBook With 0.08% Speaker Confusion — Binary_orchid · 2026-10-22
- Fireworks trains Meta's Muse Glimmer on Kalshi outcomes: +11.2pp accuracy over Jev — sophiamyang · 2026-10-22
- Gemini CLI PR reveals unreleased gemini-4-argon model with 1M context window — mr8lu · 2026-10-22