Can you provide sources of your claim that LLMs “model the world”.

You are right that it is a bold claim but here is a relevant summary: https://en.wikipedia.org/wiki/Stochastic_parrot#Interpretabi...

I think "The Platonic Representation Hypothesis" is also related: https://phillipi.github.io/prh/

Unfortunately, large LLMs like ChatGPT and Claude are blackbox for researchers. They can't probe what is going on inside those things.

It seems somewhat obvious to me. Language models the world, and LLMs model language. If A models B and B models C then A models C, as well, no?