LLMs produce output based on stuff that humans wrote, and humans reduce the world to stereotypes often. So why should this be surprising?