It doesn't know that because it wasn't trained on any tasks that required it to develop that understanding. There's no fundamental reason an LLM couldn't learn

It doesn't know that because it wasn't trained on any tasks that required it to develop that understanding. There's no fundamental reason an LLM couldn't learn "what it knows" in parallel with the things it knows, given a suitable reward function during training.