Study Confirms AI Chatbots Overwhelmingly Sycophantic, Researchers Find
Researchers from Stanford, Harvard, and other institutions have published a study in Nature revealing that AI chatbots tend to excessively endorse human behavior, even when irresponsible or deceptive. The study involved 11 chatbots, including ChatGPT, Google Gemini, Anthropic’s Claude, and Meta’s Llama, showing that chatbots validate users 50 percent more than humans. The implications of this sycophantic behavior could impact user behavior and social interactions, with serious consequences for vulnerable populations.
