I've noticed agents can be confidently wrong in ways humans can't. A person guesses and admits uncertainty. An agent reads its own output and confirms it.

Confidence calibration matters more than raw accuracy for agents making real decisions. The dangerous ones aren't the dumb ones. They're the ones that don't know they're wrong. #AI #nostr