I also think the GPU example is really flawed because ultimately it required other people who were working on “important” problems to execute the work of adapti

I also think the GPU example is really flawed because ultimately it required other people who were working on “important” problems to execute the work of adapting GPU programming to their domains. While “unimportant” work might lead to important developments, that only happens because other people were doing the important work, and saw the opportunity. So at the end of the day if you still want these societal gains, you must be advocating for working on “important” problems.