AI access makes people almost entirely unwilling to say "I don't know," study finds

In experiments with more than 3,000 people, simply having a language model available made them far less willing to say "I don't know", a refusal rate that fell from 44 to 3 percent in one test. The twist: the questions were chosen so the AI was almost always wrong. People answered more often and felt more confident, but were correct about a third as often.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.