Comment by zmj

It's not contradictory to say that expertise is a multiplier, and that models are systematically underconfident in themselves.