"Map difficulty breadth-first before diving deep Tao draws on AI's performance on the Erdos problems to articulate a new research strategy: let broad, moderately capable AI tools "map out and clear out all the easy observations" across an entire field first, and then "identify certain islands of difficulty which human experts can come and work on." Previously, science was structured around depth because "humans can"t do breadth," but now that breadth is cheap, the productive order is to survey widely before committing depth. Anyone managing a research agenda, product roadmap, or learning plan can apply this: do a low-cost broad scan first to locate where the hard problems actually are before allocating deep effort. Source: Terence Tao, Dwarkesh Podcast, Terence Tao – Hardest Math Problems, AI"s Limits, and Scientific Progress"