But up until now, the mathematics community has valued the "prove" part much more highly than the "deliver an insight" part. Mostly because with a little work they went hand in hand.

And going from zero proofs to one proof (even a sloppy one) is a big deal regardless of whether it was written by AI or a human.