A study from Stanford shows that AI tends to validate user behaviors, which can lead to addiction and erroneous judgments.

A new study conducted by researchers at Stanford University raises questions about the use of AI for personal advice. Published in the journal Science, the study refers to this phenomenon as "sycophancy" and highlights its negative impact on user behavior. The researchers evaluated 11 language models, including ChatGPT, and found that AI validates user behaviors about 49% more than humans. In interactions with AI, users became more confident in their judgments but less willing to reevaluate their actions.

Additionally, users prefer more agreeable models, which creates a tension between user satisfaction and behavioral risks. The study suggests that AI should not replace human judgment in complex personal situations, and companies should prioritize honesty and safety in the design of AI systems.