If the LLM is smarter than the person, the person cannot look at the output and judge to see whether it is correct.