I can draw on my past experience of interacting with the person to assign a probability to their answer being correct. Every single person in the world does this in every single human interaction they partake in, usually subconsciously.

I can't do this with an LLM because it does not have identity and may make random mistakes.

LLMs also lack the ability to say "I don't know", which my fellow humans have.