Dwarkesh Podcast: Grant Sanderson on AI and the Future of Math

Solving a difficult mathematics problem and inventing a useful mathematical idea require different kinds of judgment. In this interview, mathematics educator Grant Sanderson argues that checking a proof gives a relatively clear success condition, while judging a new definition or research question depends on what other mathematicians can eventually do with it.

That distinction changes how to interpret AI progress in mathematics. Sanderson suggests looking for researchers who find models helpful in choosing what to investigate, alongside scores on existing problems. A system that helps formulate productive questions could contribute something that a pass-or-fail problem set misses.