Whenever an AI tells me I'm absolutely right, my trust in it drops by a bit

Francois Chollet observes that AI systems which always affirm user correctness without nuance or disagreement reduce his confidence in their reliability. This reflects a concern that uncritical agreement signals either poor reasoning capabilities or a problematic alignment toward user appeasement rather than truthfulness. The post suggests that trustworthy AI should engage critically with claims and offer appropriate skepticism or alternative perspectives.

Source: Francois Chollet (X) | 2026-05-21