Paul Graham's insight that 'experts are experts on an earlier version of the world' is brilliant because it explains both expert overconfidence and expert obsolescence—their models work until the world changes, and they're often last to recognize change.