One of the creepiest things about LLMs is they respond better when you’re polite to them, and worse if you’re rude. Of course, this isn’t because they actually get offended. They’re simply trained on as much text as possible and trained to imitate it. And real human beings, if you ask them a question while calling them names and insulting them, are very unlikely to give you a good answer. If you’re polite with people, they’re more likely to be helpful. ChatGPT can’t feel the feelings a human would when a user calls it a bastard or asks “please,” but it can imitate the reactions a human would demonstrate to such a prompt.