 Great. Let's start today by briefly thinking about what we learned last week. So first off, wow, last week was pretty crazy. It was crazy for us organizers, scrambling to get it all in line. It was also crazy for you. And in a way, it was designed to be crazy for you. What we wanted to show you is how the overall thing works, like a deep learning system in the right environment and so forth. And it's very early for that. Now, we haven't shown you the pieces. So it must have felt like we all want. And that's okay. I promise, from now onwards, we build step by step towards the goal of really understanding deep learning. We get to see all the tricks that jointly make alpha zero work and can beat all of us in just about any board game. Now, like, keep in mind the generality of what we just saw. This is a system that can be adapted to just about any board game and beat humans easily. What did it have? It had an artificial neural network to estimate value and policy, both at the same time. Around this kernel of deep learning, it had a Monte Carlo research system that was used both for playing itself and for learning how to play. So let us briefly stare at the ANN code that made all of this possible. What we see here is we have an initialization that builds it. It has dropout and we'll get to what that means, conf layers, fully connected layers and outputs, namely, value and policy. And then it has the forward code that does the actual computations. Now, where we take the input, we have a dropout, relu, conf net, two layers of that and then it has two layers of fully connected and then it has the outputs with their relevant nonlinearities, tangent, softmax and we'll get at all those details what these words that I just used mean over the next couple of weeks. So it's truly impressive to see what the system can do in practice. So here's one of the moves of alpha zero. Look at this bot. It did G4. Now keep in mind it's playing with white. If you're simple-minded with a human-made policy, it looks pretty bad for white. Not like Black has more material at this point of time. But look what happens after G4. After G4, Black basically can do absolutely nothing. It can move on A or it can move on C, which just immediately leads to it being removed the material. It pretty much can't move anything on the right-hand side because it would immediately lose significant amounts of material. So it's a rather desperate situation and it's great to kind of see the moves that alpha zero taught itself and in fact it's having major impacts on the field of chess playing at the moment where people have these super strong computer systems now which change how humans themselves operate in those kings. So what we want to start the week with and we will start every week with that is I want you in your part to discuss what you learned last week, to discuss which things you didn't understand last week and which things you hope that you will learn yourself or in this course.