For answering questions, LLMs could be helpful for finding a needle in a haystack, but for anything important any positive return needs to be verified, and any negative return should be ignored.