As the other poster said, they can check themselves but this requires an iterative process where the output is fed back in as input. Think of LLMs as the output of a human's stream of consciousness: it is intelligent, but has a high chance of being riddled with errors. That's why we iterate on our first thoughts to refine them.