Dwarkesh Patel Argues Mathematics' AI Avalanche Is Coming for AI Research Itself

Dwarkesh Patel argues that AI's leap from failing competition math to solving a Millennium Prize problem in 88 hours is a preview of recursive self-improvement in AI research, which — unlike pure math — targets empirical efficiency gains that current systems' pattern-matching strengths can readily unlock. He cites Anthropic's claim that its rate of progress on well-scoped tasks is now 2x what it was a year ago and predicts top-human AI researchers by 2027-2028, superintelligence by 2030.