Periodic Labs' Fedus and Cubuk argue AI materials discovery needs physical experiments

Periodic Labs co-founders Liam Fedus and Ekin Dogus Cubuk, in a Latent Space interview, argue that AI scientists cannot rely on more data alone, because scientific discovery differs from math and coding and experiments remain the ground truth.

They describe reinforcement learning grounded in physical experiments, AI-driven materials characterization, and the view that failed experiments can still be valuable training data.

The transcript was truncated before the discussion of giving lab instruments a "140 IQ" was completed, so the report does not show how that idea would work. The account is the founders' own description of their vision, not independently verified results.