Someone on HN once said “It’s hard to tell people the solution to a problem they don’t have”. Building software that solves a problem that the intended users don’t understand that they have is also a real problem.
As for building a solution to "a problem that the intended users don’t understand that they have", I'll agree that this may not be a good thing for a startup to do - but when I see how basic research is being defunded and drowned in red tape, and people saying that startups are the new, distributed R&D departments, I feel the middle has fallen out. When startups do a greedy gradient descent, and academia struggles doing science, who's going to work on solving problems that require solutions one or two steps beyond what the intended audience can understand? FLOSS community? What about outside software?