 And now, simply load a pre-trained one so that you have how this looks like after training. It trains well enough. You can see that the curves converge very nicely, but is it any good? It sure is good at predicting visits and at predicting value. Now it is actually very good, and we will be able to test that. Now this is amazing. We have AlphaZero, it solves Othello, it could solve many other games. Now like, keep in mind that when we constructed this, there was nothing specific about Othello that goes in there. No, it could be any other game. It could be a real-world situation. It could be something in biology, it could be something in energy, it could be something about the money that is made by a company. In general, we now have a system that solves such reinforcement learning systems. And in a way, it's just so cool that we can get at these very cool systems that solve such problems within a relatively short amount of actual coding. And in a way that we will be able to understand all the components. And you will all get out of this deep learning course that you have an intuition for how exactly every component that we use today, how exactly it works. This is generally how neural network approaches when they're good come together. We have a network or two, depending on how you think about it, that deal with policy and value. We have a tree search, a symbolic system that goes around it. In lots of cases, we will find that deep learning itself isn't enough. We need to combine deep learning with approaches that are outside of deep learning itself. And in the code, we used many of the cool codes that we have our disposal in the domain of deep learning. And if you have time, go play the agent. It will in all likelihood win rather handsomely against you. And if it doesn't, well, give it a few more hours and try it again. So if you have time, play the agent and thanks so much.