[klyrs]

Compare the size of today's training sets (unknown, I'm guessing PB given who's doing the work), models (TB for the good ones), versus the sizes of quantum computers (hundreds for gate model, a few thousand for adiabatic). We're looking at about 10 orders of magnitude (30 doublings) of progress that would need to be made, for quantum + ML to make any sense purely on dimensional analysis. If we're generous and give both sides a doubling every year, that's still 15 years out. My money is on classical computing owning this space for a long time.