If you don't just move on to another problem you will need people who know how to push these systems further.
If you don't just move on to another problem you will need people who know how to push these systems further.
I am rather newbish in the whole space, but most everything I've seen is quite often just knowing enough to turn the right dials the right way. And I do wonder how long will it be before some of these dials can start turning themselves with more iterations with a simple, "yes that's what I want, or no that's what I don't want"
Sure, it might take some more cycles for it to find optimal in this example.
Understanding what models are doing and adding new features/data transformations is standard work for ML practitioners now. If you believe the deep learning hype, architecture engineering is the new feature engineering, and to the extent that that is true, industry will essentially resemble academia to some extent. Maybe not making anything groundbreaking, but mixing ideas together.