Stanford Report ; AI overly affirms users asking for personal advice: Not only are AIs far more agreeable than humans when advising on interpersonal matters, but users also prefer the sycophantic models.
"Researchers found chatbots are overly agreeable when giving interpersonal advice, affirming users' behavior even when harmful or illegal.
Users became more convinced they were right and less empathetic, but still preferred the agreeable AI.
Researchers warn sycophancy is an urgent safety issue requiring developer and policymaker attention."
No comments:
Post a Comment