I thought this was a really interesting presentation. Sangbae Kim (MIT), describes the challenges still facing robotics researchers as they try to build robots with more human like abilities. Given the astonishing leaps LLMs have taken, one might expect that physical action might also be "tokenizable". Kim pours a bucket of cold water on that idea.
Stanford Seminar -- Physical Intelligence and Cognitive Biases Toward AI
Stanford Seminar -- Physical Intelligence and Cognitive Biases Toward AI
