While I agree that the quotes you made sound more like marketing than science, antirez has a point here: They might be using GA to evolve a new activation funct

While I agree that the quotes you made sound more like marketing than science, antirez has a point here: They might be using GA to evolve a new activation function represented as a GP tree. But that's just my guess. And even so, I am not terribly convinced that would lead to astonishingly new activation functions that are better than the known ones.