You made a very specific claim that random fluctuations could have the same effect as adversarial examples. I was addressing that.