>Given any particular example, LLMs might hallucinate more and a human might do a better job at accuracy

This drastically depends on the example. For average trivia questions, modern LLMs (even smaller, open ones) beat humans easily.