Martin Wattenberg: Models within models - how do LLMs represent the world? | Berkeley RDI Center on Decentralization & AI | Podwise