 So what we saw is that ReLU allows us to approximate any function and ReLU, if you think about it, it's super easy. Now I can take that the output is the input and if the input would be smaller than 0, I will just clip it to 0. But, wait, wait, wait, when we implement a function in deep learning that is used a lot like the activation function, then how fast it is actually really matters. And you will see it can easily make the difference between something that runs slowly and something that runs rapidly. So I want you to just see how it matters and I want you to try different ways of implementing ReLU and compare the speed of them.