my view is that inference will eventually (probably quickly within the next 3 yrs or so) move from centralized data centers to local devices as chips become more powerful and models become smaller and...