I actually mostly agree with this -- the points of contention might be how we each define "widely used" and "eventually". Without those parameter settings, the statement is basically un-falsifiable. But still I think he's saying something important, and why MoE can't be ignored.