Galton's Law of Mediocrity: Why Large Language Models Regress to the Mean (arxiv.org)