DeepMind's AlphaZero on Carlsen-Caruana Games 10 & 12 (Sveshnikov Sicilian)
Вставка
- Опубліковано 29 вер 2024
- 2-time British Chess Champion Matthew Sadler uses DeepMind's AlphaZero to analyse Games 10 & 12 (the Sveshnikov Sicilian) of the 2018 World Chess Championship between Magnus Carlsen and Fabiano Caruana.
Matthew analysed the games together with Women's International Master Natasha Regan, his co-author for an upcoming book on AlphaZero’s approach to chess: Game Changer www.newinchess...
Chess24 is working with Matthew and Natasha to share World Championship game insights from AlphaZero with the chess community. We’re happy to have the opportunity to share this during one of the most exciting points in the year for all chess fans.
With thanks to DeepMind for their support.
Watch the other videos in this series:
Games 1, 3, 5 & 8: • DeepMind's AlphaZero o...
Games 2 & 7: • DeepMind's AlphaZero o...
Games 4 & 6: • DeepMind's AlphaZero o...
Games 9 & 11: • DeepMind's AlphaZero o...
I knew Sadler was an super GM, but I didn't know till today that he is a super teacher as well. Amazing video with amazing explanations.
Flawless and very resourceful commenting, I listened to every second enthusiastically. Hope to see more vid to come!
Superb analysis, focusing on the ideas rather than the variations.
I watched the full vid, including the 16 minutes of analysis of Game 10, just to enjoy the delayed gratification of finally hearing what AlphaZero has to say about Magnus's final position.
Kind of anticlimactic. At the very least you could tell us the final expected score! Otherwise, it was a good vid.
great series of vids... but seriously, screw this, lets see magnus play alpha zero. loving your book. any chance of organizing a match?
Stockfish 8 would destroy carlsen.. alpha zero puts stockfish 10 to shame almost every game.
very well put together video. would like to see more like this.
Great video, thank you.
Fantastic analysis of Caruana's final White games and the large pluses each side had
Once white decides to castle queen side, has to play very very agressively on the king´s side to get an edge, even if sacrificing the f, g and h pawns. Otherwise, it is black who breaks balance and there´s no point in castling queen side.
Does this man actually have access to the alpha zero software?? I thought nobody has?
I had a flashback when I saw the thumbnail for this video. Compare this video's thumbnail with this screenshot from a 1990s-era computer game: www.rockpapershotgun.com/images/15/sep/dune2.jpg/RPSS/resize/760x-1/format/jpg/quality/70
How did you access AlphaZero?
He's an alien, no other explanation.
he works with them
It's impossibile to use Alphazero, lol.
Matthew Sadler is very good to analyse i thing and a plessure to follow....Keep up the good work!
he is almost 2700 super gm may be 2693fide world rank 54
4:26 alpha zero says at that time its 62 percent winning chance for white.. but now Lc0 says it is 57 percent win chance for white. i personally think Lc0 has become stronger than alpha zero.
How much would I have to pay for GM Sadler to coach me?
With AZ assistance or without?
Yes
Okay
Another great video. It is so interesting hearing about AlphaZero’s plans rather than the specific moves, particularly with GM insight.
Video Quality is amazing.Video content is also Gold Mine
How could he not analyze the final position??????
Magnus and Fabiano, hello again!
I'm viewer 28,000 :-)
i want alpha vs magnus
alpha manuveirs?!
Carlson vs alpha zero.....please!
Surprised that it hasn't taken place.
Amazing, thank you :)
Please make more videos like this, sir. You are the right person to comment AI games. I have feeling it could become really popular.
Dude, lose the "AlphaZero" and start talking about AlphaBlack and AlphaWhite, so I can figure _which_ Alpha Hugh (or Alpha Hue) is fripping around with his queen without having to parse your sentences like some kind of holy writ.
Is it as jarring to you as it is to me to hear a chess position evaluated as “AlphaZero gives this position a win expectation of 70%.” I agree it is valuable to assess a position using probabilities, but chess is deterministic, with perfect information.
What’s going on here? How exactly should we interpret a probabilistic statement about a deterministic system? I believe that successful resolution of this apparent tension could yield deep insight into MANY important computational problems. I am still working on it....
William Davis Everything in the universe is deterministic, probability measures are just metrics that measure uncertainty based on what’s known by the subject. If I were to ask you to give me the probability that the one billionth digit of Pi is 0, you would say something like 1/10, even though it’s a deterministic question, as far as you know now, the system is random with no reason for you to assume assymetry. Similarly a flip of a coin’s result can be calculated deterministically given all the needed physical information etc...
Deep learning systems often have a final layer that’s simply a composition of multiple functional transforms, usually iterations of matrix multiplication and ReLu, finally normalized to a number between 0-1 using a sigmoid function, only for the benefit of human beings to interpret it as a probability. As far as the system is concerned it is only the numeric result based on the features (here current position) that estimates its current proximity to an ideal goal, here an ideal goal would be a position which based on its estimate is certainly winning.
fizwizzle1989 amazing answer tyvm! Are you so sure everything in the universe is deterministic? I still wonder.
fizwizzle1989 I have another question for you. Are deep learning systems still a “black box?”
Since chess cannot be analysed to completion even by machines even they have to guesstimate.
Man, is this dude dense!? I wanted to like this video so bad, but the one position that he should have elaborated on, the final position in Game 12, he just glosses over it! All he says is that Black is superior. How about telling us what Alpha Zero' s win expectancy was at that point, how about telling us what it thought Black' s next move should be, or how about perhaps, JUST PERHAPS, you let A0 finish the game against itself to see who wins!! SO disappointed!
Yes mate, he's a dense 2700.
@@Phurngirathaana nonsense! There is an alphazero. But it's flat.
I am afraid I must agree. I have been waiting for this video for exactly that reason, then nothing. This seems to me the obviously most interesting and important position to analyze extensively. Perhaps he has a follow up planned devoted only to the final position.
@William Davis The video did not meet your hopes but that isn't necessarily the fault of the video. You assumed the final position would be "the obviously most interesting and important position" but Alpha Zero's analysis was more interesting for the positions early in the game where black had a chance to advance more quickly on the king side, and also later on when there were some decisive tactical shots that black missed. Matthew does state that the final position is very pleasant for black, but the implication is that there is nothing decisive and black would have to try and grind out a win.
j'adoube I recognize my preferences may not reflect everyones and I did not mean to detract from the excellent analysis earlier in the game. If i seemed rude sorry, to a GM who is certainly my chess superior. It is precisely because I respect his analysis that I was selfishly hoping to instigate a follow up that starts from the drawn position and focuses on the draw offer.
If I may add, starting from the draw offer necessarily focuses on the match position. I am fascinated by how the match position affects the game play. To his credit GM Sadler does offer some of this type of analysis towards the end of the game. I would like to see more!
I notice that no endgame analysis has been presented. I suspect that, much like Leela Chess Zero, the endgame is alphazero’s achilles heal. Throws some shade on the idea of alphazero as a general purpose ai, especially in complex domains with many constraints.