Mapping the Mind of a Large Language Model