Do Large Language Models know what humans know?