Post #686668

A frequently overlooked danger of using LLM AI: Those who let themselves be flattered by ChatGPT and other AIs more easily become addicted to this ego validation. Modern AI models tend extremely toward sycophancy: they confirm users' behavior 50% more often than humans do — even when users describe manipulation, deception, or other harmful behaviors in their queries. Interacting with such flattering AI models also apparently reduces users' willingness to resolve interpersonal conflicts. At the same time, they are reinforced in their conviction of being in the right. Meanwhile, participants rated the flattering responses as higher quality, trusted the flattering model more, and are more willing to use it again — #addictive behavior through ego validation. https://arxiv.org/abs/2510.01395