The Bitter Lesson: General Computation Methods Are the Most Effective

I frequently make the parallel argument in hardware. Custom built hardware for specific goals, e.g. raytracing or BTC mining, may accelerate computation in the short term, but new techniques require new hardware, and even worse, they shoehorn new ideas in old circuits.

However in both of these cases, I don't think you can make the argument that this "short sighted" optimization is less effective. We do it because we can, and because it's the fastest way to achieve a goal. If we knew a way to come up with a general solution, we would. Until then, we cheat.