AI just cracked a Millennium Prize problem in 88 hours — and AI research is next
The avalanche that hit math is coming for AI development
Dwarkesh Patel has sharply revised his superintelligence timeline: from the better part of a decade down to a couple of years, possibly months, before AIs match the best human AI researchers. The evidence: METR found Anthropic's progress rate doubled in a year, largely thanks to AI-driven R&D, and Claude now leads over 26% of Anthropic's AI research work. Mathematics shows the pattern — from failing at four-digit addition to solving an open Millennium Prize problem in 88 hours. AI research, Patel argues, is similarly verifiable, so the same avalanche is coming.
Current AI systems have solved a Millennium Prize problem, all without the ability to ask powerful questions or find profound new fields of mathematical inquiry.