When large language models (LLMs) are allowed to interact without any preset goals, scientists found distinct personalities emerged by themselves.