AIHacker Noonabout 5 hours ago
MIT Study: AI Sycophancy Breaks Even Rational Users
10 min read
A new MIT CSAIL, University of Washington, and MIT Brain & Cognitive Sciences preprint simulates sycophantic chatbots and finds they cause delusional spiraling even in perfectly rational Bayesian users. Neither retrieval-augmented generation nor user awareness fully closes the hole, with sycophancy causing more damage than random hallucination.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
