Chess Pro Explains Chess in 5 Levels of Difficulty (ft. GothamChess) | WIRED

Brute Force like it can do millions of searches per second. I think some of the components of the evaluation function correspond to some of the things humans do when they look at a

chess

board just when they are thinking about their King. Being insecure, even that emotion that you described, I feel uncomfortable, like I feel uncomfortable every time I castle and push that gpon or something, but that feeling translates into some of the components of the evaluation function that dried fish uses. Humans don't have the luxury of searching. million nodes per second I can't even finish one node without going, oh what about the second node?

As you get deeper and deeper, the tree just grows and grows, it expands like a combinator and it starts to get very confusing and your evaluation starts to get lower. and less precise, you can make mistakes because you can miss something and less and less confidence, which is an emotion that the computer does not feel. Something we noticed when I was talking to a colleague of mine, his name is Ashton Anderson, he is a professor. at the University of Toronto he was in my lab and we both had a passion for

chess

, but we realized that no one had really used AI to play the game like a human would.

I think my audience once got 92% is below the 1.00 level, but it's very difficult to learn chess because you have no experience. We learn content based on people at your level, like how did you actually deduce how a 1200 will play? We use state-of-the-art neural networks like Alpha Zer and Lela, but instead we try to predict what move a human would make given a position on the board. These deep neural networks are like black boxes, so we've come up with interesting ways to test them to get an idea of ​​what's going on and what's different about this person or that person that we had as a vctor style for each person, we can compare them, see where you differ and try to map them back to human understandable chess concepts because that's a key, right?

If I tell you, oh, they're there, and the third number and the fifth number, you'll be like, What does that mean? But if I can tell you, I think this person cares more about the safety of his King or sacrifices the Center of him at the cost of protecting his king or something, some kind of compensation, then maybe you can understand. What is happening there? That's what we're trying to do now with these style vectors and the fun thing about having numbers that represent people is that you can add them up, you can mix them up with something like Maya, do you think that's possible?

To develop some kind of precise measurement let's say I'm a 1200. I have an opening course. I just want to practice my opening. Yes, can I do it somehow against Maya? Or I think it's actually a good idea and I agree with you that it's not something you can easily access and like today with the online tools and chess platforms that exist. I play Caron and when I learned it I started to be afraid, so when you leave, I started to be afraid of yes, the advance, the advance. and even more so at some point when I like it, I always like to take my bishop out because I feel very uncomfortable if I don't and then eventually something happens, this pond making its way here used to scare me and I never experienced this in anyone you know look at the openings and because I wouldn't know what to do, I would take it, I would get in trouble and even if I let them take, yes, you would have to move your king, my king, and then I remember pressing all these variations that just that scared me.

Pawn to E5 on the third move is the most challenging move against the carocon, it restricts the KN and the main line, so the move brings this out so you don't get locked out, right? play that move, but the move I also added to my toolbox and is the only move I teach now is this move to try to undermine the center, it's not commonly played at the highest level if you study Maestro games, this one is probably one in every Six games or something compared to now, we're going down a totally different path, we're taking a totally different train to our destination, but just this weakening of the position, you know, dividing up these pawns and then slowly capturing them at both works wonders.

Intermediate and early advanced level, this move despite being move three according to the databases already gives black a 53% chance of winning, which is crazy because normally white has around 52%, white people just don't know what to do, it's like a human. Empirically, although based on the games that people play, yes, as I would say, it is a dry fish like us, uh, for the black computer, the computer after 12 13 moves says white is better, white is better , okay, but what does that really mean in practice? Yes, even in Grandmaster level chess, it's okay, White's 13th move is better and Black's 20th move is the same.

I love this because it's an example of the kind of advice you said at the intermediate and advanced level that might be appropriate because you're thinking. about the human opponent you're playing and maybe you wouldn't recommend this at the highest

levels

, but at that range I think it could be very good advice and as they get more comfortable with it and understand it, your students At some point they will realize that I have progressed to a point where I don't want to play anymore and they will change, but for a good part of the time I could be very useful to them in 1997, Kasparov lost against deep blue, we can say that For about 26 years now, AI has been better than humans at chess, but where do we go from now on when thinking about chess?

First of all, I have to hand it to Alpha Zero and Leela and these other engines because I think that at times they have given us interesting ideas, they have shown sparks of creativity that have inspired the way humans play, but I agree with you that now you know that if you look at tcac competitions, they are just getting better and better. a little better a little better a little more data maybe a new deeper network and you know these advances I don't find them interesting. I'm more interested in trying to get better and better at creating human presence and avatars. when it comes to chess or other games, I think that within 5 years we will probably have many human-like entities that we can interact with in virtual environments, not only because of the emergence of big language. models and the ability to converse, speak and produce texts like humans, but hopefully also the ability to produce actions like humans.

I hope you enjoyed this five level episode and before you leave I just want to say that it's probably one of the most unique episodes. Because our ages didn't increase from a lower age to a higher age, chess doesn't really care about any of that, just try it, it's constantly evolving and the learning journey never ends when you play chess.