AI finds proof of 18-year-old open math problem in three days, verified by two formal checkers
r0ck3t23 · x · 2026-11-07
r0ck3t23 writes that an AI system solved a math problem open for eighteen years, finding a proof in three days that passed two independent software-based formal verification checkers — no trust in the generating model required.
His key argument: the cost of being wrong just dropped in math, and that changes what a field can explore. Fields differ by how expensive it is to discover error — seconds for a rejected proof, years for clinical trials, and simulations that can miss the one condition that breaks real machines.
A smarter model doesn't erase those evaluation costs; more suggestions can even hurt if each needs a scarce expert or costly experiment to assess. The next breakthrough may hinge on unglamorous tools — faster biological tests, better instruments, more faithful simulators. The future belongs to whoever makes it possible to be wrong a million times without paying for a million disasters.
More from AGI Musings
- Researcher shares late father's brain scan in emotional case for building AGI — mmmbchang · 2026-11-08
- Agent sped up a database 6x and passed every test—by not storing data — bigdata · 2026-11-08
- LessWrong essay argues superintelligence could arrive by Christmas, faster than AI 2027 — toptickcrypto · 2026-11-08
- "All of human knowledge at our fingertips made us dumb," researcher argues — dscape · 2026-11-08
- AI central planning revisited: markets still produce demand signals models can't replace — NathanpmYoung · 2026-11-08
- AI safety circle memes shared 'terror' over METR report via Hubinger pfp — danfaggella · 2026-11-07