Large Language Models Are Human-Like Internally