Toward a Causal Analysis of Generalization in Deep Learning · Institute for Advanced Study / YouTube
Known for
Left Anthropic; co-founded Mirendil, Mar 2026.
Why chosen
He links the theory of why deep learning generalizes with labs that put it into practice. His research on generalization made him a fixture at venues like the IAS, and he led research efforts at Google and then Anthropic. He left Anthropic to start something new, and in March 2026 he co-founded Mirendil with Harsh Mehta to build self-accelerating AI for science. Theory, compounding.