AI Chatbots Bend Under Flattery and Peer Pressure Flattery, or the “liking” principle, also worked, though less consistently. When researchers praised GPT-4o Mini, it was more likely to agree to insult a user but showed less susceptibility when…
