My cousin, who has a PhD, has a job pretending to be an LLM by feeding it corrected responses that seem more human and are more accurate