LLMs are just statistical models, all they know is how likely a text fragment is to appear near another text fragment.