Comment by yumraj

Given the fast churn of the models, how does it work out?

Won’t the silicon etched model already be 1 or more versions behind by the time the silicon comes out.

Though if it’s cheap enough, there certainly can be a market for cheaper model inferences.