Lots of excellent references showing LLMs have internal models. Good precis of the philosophical argument that meaning doesn't require reference.