303 | AMA | February 2025

10938.413 - 10964.094 Sean Carroll

The large language model is optimized to give sensible-sounding answers to human beings asking questions of it. Now, it may very well turn out that in that black box of many, many layers of deep learning, the way the LLM does that is essentially to invent intelligence, to invent a model of the world, invent sort of counterfactual reasoning, invent all those things. I'm open to that possibility.