> If it takes a few thousand pages of textbooks or other reference material to gain competence

This is a huge assumption and not one I’m sure holds up. In my experience gaining competence is often more a matter of hands on experimentation and experience, and the thousands of pages of reference material are there to get you to the point where you can start getting hands on experience, and debug your experiments when they don’t work. If AI can meaningfully cut back on that by more efficiently getting people to the experimentation stage, it absolutely will be more effective. And so far in my limited experience, it seems extremely promising.