"Taking things apart" is a very useful framework for understanding AI. How do we resolve the (seeming) paradox that some skilled workers report wonderful results from their work with AI, while others are full of dire warnings about the lurking defects in their AI-assisted outputs?
3/