As far as I know we don't know much about metacognition in LLMs, though? Not sure