Mathematicians need to stop toiling on proofs and acquire taste for meaningful questions.
Right now, people graduate top of their Harvard class and work on a question that was invented 150 years ago and everyone just assumed to be true. Kind of meaningless, right? Could have just waited three years for a full AI-only proof without human intervention.
And you cannot acquire taste without getting dirty.