If you had to project what kind of pattern we'll see in inference cost v.s. time, what would you predict? Over the next 2, 5, 10+ years?