303 | AMA | February 2025

10989.405 - 11012.545 Sean Carroll

It's not that LLMs don't become better and better. They're clearly becoming better and better. But they aren't perfect, and they make mistakes, and they make failures. And my point has always been the types of failures they make are precisely the type you would expect if they were not being real human intelligence, if they were not causally mapping the world and inventing counterfactual reasoning.