“Dario Amodei predicts a future where AI models will make mistakes much less frequently than humans, but the mistakes they do make will be stranger and less intuitive to users.”