This episode explores why materials discovery cannot be solved by reasoning over existing knowledge alone. Periodic Labs describes an end-to-end loop of predicting, synthesizing, characterizing, and iterating on materials, supported by reinforcement learning grounded in physical experiments. The conversation examines uncertainty, noisy measurements, phase identification, density functional theory, simulations, multimodal characterization, laboratory automation, negative results, and training models on the process of science. The guests argue that future frontier models will still need experiments because discovery concerns what lies beyond their training data. They also discuss intelligent instruments, custom laboratory hardware, semiconductor deployment, commercial outcomes, and how automating thousands of experiments could expand the search for superconductors and other transformative materials.