AlphaZero goes fishing with Stockfish
HTML-код
- Опубликовано: 16 окт 2024
- It's another Queen's Indian Defense, but with a twist. After AlphaZero undermines f5 with 10. d5, the king knight does not look to occupy the f5 post, but rather g5 to set up a fishing pole tactic. A lot factors must be in place for the fishing pole to be effective. Namely the white queen and king rook must observe, however indirect, the h7 square. What does this imply? Well, routine moves like 0-0 by AlphaZero will not be executed too soon, if at all!
PGN:
1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. g3 Bb7 5. Bg2 Bb4+ 6. Bd2 Be7 { E16 Queen's Indian Defense: Riumin Variation } 7. Nc3 O-O 8. Qc2 Na6 9. a3 c5 10. d5 exd5 11. Ng5 Nc7 12. h4 h6 13. Nxd5 Ncxd5 14. cxd5 d6 15. a4 Qd7 16. Bc3 Rfe8 17. O-O-O Bd8 18. e4 Ng4 19. Bh3 hxg5 20. f3 f5 21. fxg4 fxg4 22. Bf1 gxh4 23. Bb5 Qf7 24. gxh4 Bf6 25. Rhf1 Rf8 26. Bxf6 gxf6 27. Rf4 Qg7 28. Be2 Qh6 29. Rf1 g3 30. Qd3 Kh8 31. Qxg3 Rae8 32. Bd3 Bc8 33. Kb1 Rf7 34. Qf2 Bd7 35. h5 Ref8 36. Bc2 Be8 37. Rf3 Re7 38. Rxf6 Qxf6 39. Qxf6+ Rxf6 40. Rxf6 Kg7 41. Rxd6 Bxh5 42. Kc1 Re5 43. a5 bxa5 44. Kd2 Be8 45. Ra6 Rh5 46. Bd3 a4 47. d6 Bf7 48. d7 Rh8 49. e5
AlphaZero: Shedding new light on the grand games of chess, shogi and Go:
deepmind.com/b...
Full peer-reviewed AlphaZero paper:
science.science...
Internet Chess Club (ICC)
Software: Blitzin
I'm a self-taught National Master in chess out of Pennsylvania, USA who was introduced to the game by my father in 1988 at the age of 8. The purpose of this channel is to share my knowledge of chess to help others improve their game. I enjoy continuing to improve my understanding of this great game, albeit slowly. Consider subscribing here on RUclips for frequent content, and/or connecting via any or all of the below social medias. Your support is greatly appreciated. Take care, bye. :D
★ LIVESTREAM / chessnetwork
★ TWITTER / chessnetwork
★ FACEBOOK / chessnetwork
★ PATREON / chessnetwork
★ DONATE www.paypal.com...
Here is the AlphaZero vs Stockfish playlist:
ruclips.net/p/PLQsLDm9Rq9bEvlHwbsWlt5sta3M2t0aGe
Hi Jerry, excellent videos and analyses. There is one game between all those 200 games that is a Semi-Slav, where Stockfish (Black) managed to get a pawn in c4 and g4, and then things get super complicated (atleast for me xD).
Alpha zero makes the most interesting plays in the history of chess!
I would LOVE to see the thousand-move games, where both computers use continuous draw threat to force moves to be made. ;) Please post one of those videos!
Thanks Jerry. I love your commentary and basically everything you do on these chess games. Are you ever going to participate in the titled arenas that we usually watch Carlson play? I’d really like to see how you do. What is your current bullet or blitz rating?
ChessNetwork 5
"I've only skimmed through over 100 of them."
Jerry, December 2018.
I believe the "only" refers to the fact that he skimmed them, and didn't analyze deeply ;)
He's slacking off a bit
Such laziness. Kappa :)
AlphaZero is busy writing "My 65535 Most Memorable Games"
Ah you also studied comp sci?
Dan Kelly What does it mean?
@@elijahmiles8874 Comp sci = computer science. 65535 (2^16 - 1) is the largest 2-byte (16-bit) integer you could store. In binary, that number is the highest integer you can have with 16 digits.
@@FiresBZ More like "My 2147483647 Most Memorable Games"
18446744073709551615
:P
Me: wow stockfish 8 sucks
Also me: stockfish 4 mated me
SpinSpider Just so you know, the “4” in “Stockfish Level 4” refers to a different number than the “8” in “Stockfish 8”. SF8 is the eighth version of Stockfish, which can be configured to play at many different difficulty levels. In this video, SF8 is playing at full strength. When you play SF level 4 on lichess.org (for example), you are playing SF10, configured to be much weaker than its full strength.
Lool
Hahaha
Lmao
underrated comment
It's interesting how Alpha teaches us how we've all been misled by our training. When we, as humans, see that we can get a rook for a bishop, it's almost an automatic move to take it in most cases. Since Alpha taught itself from scratch, it was never miseducated into thinking that any piece has a specific set value. The value of any piece is entirely dependent on what it can actually do on the board, given the position of all the pieces in the game, and has absolutely no inherent value otherwise.
I love the games where it's like
Stockfish: Ha! It's the mid-game and I'm up three pawns! Suck it, AlphaZero!
Alpha: Material means absolutely nothing. Look at the board. You're losing.
Indeed, but if we're trying to reduce inaccuracies in general, it's worth taking note of what AZ seems to show - that thinking of pieces as having set values is itself an inaccurate and unnecessary way to analyse a game. The value of a piece depends on what it's doing now, and what it has some prospect of doing (not in principle, but in the actual game unfolding). To some extent, we've always known this, else such things as sacrifices would never have appeared in human GM games to begin with. AZ is just able to make much better use of this higher-resolution thinking than we can, for the reasons you give (I don't think an non-augmented human is ever going to beat Alpha Zero!).
Humans thought they had mastered chess but engines and AI proved otherwise.
Except at 12:59 Stockfish 8 recognizes gxh4 (the move played by AlphaZero) as the best move instead of bishop takes rook
@@MrSupernova111 Was it against a chess engine or AI?
I always thought the point values are just for beginners to understand exchanges.
Even intermediate players understand that pieces value is dependant upon what they are doing. It's engrained into you once you understand the value of connected 7th rank pawns can be more powerful than a queen or the first time you get mated by minor pieces while your up material
Looking at high level games, tals games come to mind specifically, they will trade down material if it is a sure advantage.
If theres no way to materialize the advantage then a rook is more powerful than a bishop but if there is a strategic reason then bishop may be better.
Alpha zero probably just understands deeper than anyone when and how the bishop can be better.
Just thinking out loud here though who tf knows right? Certainly be not me
*Watches AlphaZero play*
Wow, these moves are so obvious. Of course you would play these moves.
*Plays against Stockfish 3*
Damn, I got checkmated in 4 again...
Story of my life lol
It's what I like to call the "wheel fallacy". Basically, something logical seems obvious once presented, such as the original man-made wheel, but whose discovery actually takes great insight to realise.
I prefer to think of it more like the Tao of Chess. Alphazero has this because Alphazero doesn't play with intentions in mind. When you have a human player, especially an inexperienced one, they'll play with plenty "more" on their mind. They may be looking to develop a particular strategy (clumsily), or they may be pre-occupied with certain pieces that are plaguing them or serving them well.
It's not just that it seems logical in retrospect, it's that Alphazero moves with what you might call grace. Once you take away all the cognitive biases that plague a human player, play seems 'obvious'.
Nah I take back moves and drag along till move 12 or 13 and then the situation is irredeemable so I close the window
@@kaniza1465 Ah, didn't know it had a term
In 2018 we have Humans playing like Engines (Caruana v Carlsen) and Engines playing like Humans (Alpha Zero v Stockfish)
What a time to be alive
In 2020 the boundary between human and engine will be no longer exist
Alpha zero does not play like a human.
Alpha Zero seems to have this incredible error free, strangling style of play. Whilst not shocking, it’s incredible to see how powerful this approach can be.
@@nofanfelani6924 people have been saying that every year for 20 years now
But what is "alive"?
Poisoned pieces, sac the exchange, gambit away a few pawns ... the machines are taking us back to the Romantic Era of chess. All 19th C players would feel right at home here!
I would, I mean they would.
🎯
Machines beating up machines. Can I tell my family I'm watching Transformers? Maybe they wouldn't give me more weird looks.
Human chess is a joke compared to this. Upload the game where Alpha Zero gives up 4 pawns to dominate with the bishop pair.
yeah that game was awesome
Anna Rudolf uploaded it on her chess channel yesterday
Yea , this one is too much 😪
@@perchix5252 link?
Coming next video:
AlphaZero goes house hunting with Stockholm.
i don't get it :( but seems funny
The pattern I see from AlphaZero.
1. Put all pieces to ideal position, than only commit to castle.
2. Do not overcommit in the center as it what hold pretty much every position. Plus, there might be chance for the king to stay in the center, keeping it active is playing with an extra piece.
3. As white, most of the time the light square bishop is shut down pretty hard. Restricting the castle at a8 further.
4. Pawn sac and keeps the bishop pair alive so there are less pawn block for greater piece activity.
5. No rush with pawn break. Force stockfisk into zugzwang, then only break open the position.
6. Last but not least, piece maneuver and space acquisition is always on point. Attacking on one side without compromising too much defending on the opposite site. It's pure masterclass.
Have to say, every game played is very impressive.
Yeah. Last but not least, for anyone who keep saying "The move is clear as daylight." Beat stockfish 8 soundly, with proof 1st and foremost of course, then I will believe you.
ok
Titles are getting better and better.
Haha indeed, hood sir.
Edit: good sir.
Haven't watched video yet but making prediction: AlphaZero turns Stockfish's light square bishop into a useless piece
Also love your vids Jerry :) you inspire me to get better at chess
Ur prediction was 100% correct;)
@@scorchedBoard well, is that even a prediction?
Fantastic, unhurried analysis.
Hi Jerry, it's everyone.
Was just watching your blitz tournaments and now im here, love your channel and you do a great job at explaining methods lower ranked players like myself can understand, thanks jerry!
More plz! Best commentary available by far. You don’t dumb it down.
The day when after 1. d4 black resigns the game is coming sooner than we thought
maybe 10 years?? 20? 960 will become the default by then tho
That's called the French defense.
XD
@@NoahAbrams01 if chess is solved and it is a win for white, either d4 or Nf3 will be the winning opening move. But like you said chess could just be a draw
Are you sure black is not the winner? First move is always a mistake and black wins is my brave guess :) So before the game starts white resigns ;)
@@Gasimoe lol maybe at 2832-2835 level
Your analysis is on another level imo.
I really appreciate your videos and hope you have more alphazero games. Break through stuff, amazing to see, and your analysis of the games is awesome.
Jerry, your commentary is so unique and fun, very instructive and covers so many variations... post as often as you'd like and you can, they're great.
Thank you, Jerry. I've been watching for years and still love your analyses. Just yesterday I wore my "Back to Tournament" white knight shirt to the gym. It's always good for a few comments. Thanks again.
210 more games!? I can't wait to watch your reviews!
i wish i lived in the parallel reality where jerry analyzed and commented and all of these :(
A hint to German terminology: The German "Z" sounds like a "tz" like in "quartz", so "Zugzwang" pronounces like "Tzug-Tzwang". Always makes me happy to see German loanwords in English though :D
T33K3SS3LCH3N a "g" proceeded by a vowel in German is also an English "k" sound, so it would actually be tsook-tsvawng.
@@wirelessbaguette8997 Very well said. You accidentally taught me why so many foreigners pronunciate the "ng" at the end of German and Dutch words as "nk".
Tsuktsuvang is the correct porrunciation 😂
Tsuk tsvang is the most correct one 😂
At 9:01 I auto assumed that "this is played with" would be followed by "a draw offer" because I've gotten so used to hearing that now.
Drawsen - Drawuana LUL
@@vgamerul4617 more like drawish giri
Everyone's jumping on these... Anna Rudolph has some great analysis too. Material is not as important as activity!
Hello, Jerry! So glad I discovered your channel here ...What I like most here is when Alfazero talks to Stockfish at the end of the game, giving him lessons on how to play. I get mesmerized of that robotic voice rebuking Stockfish...I feel like ET came back to play chess with us, humans.
I have found your analysis of alphazero games to be most useful amongst all the other ones on the internet. You elucidate positional play and strategic ideas brilliantly. I hope you keep them coming, I most certainly will be looking forward to them.
that terminator thing at the end freaked me out LOL
It would be nice if you showed evaluations for the newer version of stockfish too, to show how it has changed and maybe some insight on how stockfish has improved.
i love love love alphazero on your channel, keep up the great work man !!
Great series. Appreciate the depth of analysis.
There goes AlphaZero annihilating the poor fish again.
The fish is shit easily defeated jajajaj
+@@christianquinones9347 Play against it yourself and say that again...
something that's exciting for me about AlphaZero is it's ability to multitask, it can send multiple different threats to all corners of the board while still being connected and without weakening itself
It does seem like the best strategy to translate raw computational power into chess wins is to construct complicated scenarios where you have as many movement options as possible (ideally) while making it hard for the opponent to have an easy strategy of their own. The more time your opponent has to waste to think of a good move, the worse the move they pick under time constraints is going to be. At least with high probability.
Thank you Jerry, I am looking forward to over 200 alpha0 games commented by you. 🤠👍
Greetings from Germany, Black Forest. 🌲
Thanks Jery, I really appreciate your time and passion for the game. Cheers, Alireza :)
Omg been waiting for months, thank you a lot for continuing this playlist
Jerry, I know I speak for a lot of people where if you were to commentate every game that was released, I'd watch! Your analysis is great man!
It was wonderfull how alpha zero didn’t had any rush in winning the game to rather ending it in a cleaner way like if it cared about it
Please make more videos of AlphaZero, I love to watch them.
Oh, yes! AlphaZero returns! I watched the older AlphaZero videos but I love seeing there's a new one! As I've wrote this comment at the beginning of the video, I look forward to what AlphaZero thinks of this match!
23:53 AH, YEAH! Gotta love AlphaZero's messages to Stockfish.
The AlphaZero talking trash is priceless!
"Stockfish resigns" 👀
There is a book coming out in January with a collection of these games and analysis. Daniel King was talking about it on his channel Power Play Chess. Wasn't sure if you would be interested or if you already knew about it. Thanks for the great easy to understand commentary. Really nice for average chess players like myself.
It's crazy how even the material was throughout, but how crushing it was.
15:11 The white pawn is blocking the h-file. The other rook is defending the e pawn in case black's e pawn is forced away by white's d pawn. By moving the h rook to the f file, the bishop is being pinned (indirectly defending the h pawn) and the rook is being put on a better file.
I really enjoyed your video here! Thanks for making it!
Another fun and educational video from Jerry. Thank you!
Great video! I hope you will upload another AlphaZero game soon :).
Jerry at 14:56 I think the reason you want to keep the D rook where it is is so that the e5 advance remains a threat. It's the only explanation I can come up with.
Zugzwang! Yay!
PS: thank you so much for covering those games. Love your style, always instructive, especially for me as low-level-layman :)
Best chess commentary on this channel. I watched all your videos
Excellent interpretation. Pretty deep analise.
Move 25 if you move the D1 rook there could be a Bishop sacrifice on D5 with a now very active black queen
I'm glad they released more games cause it means more videos from Jerry!
Hey Jerry, loving the content. Could you please cover the Tal vs Botvinnik match in 1960? Thanks in advance.
Beautiful games, thanks for sharing.
I would take the 'tale of the tape' with a pinch of salt as it only analyses each move for a fraction of a second.
Neural networks is an interesting technology to follow. Leela being one of them.
Jerry you should go fishing again sometime, those chesscube videos where you go fishing pretty much every game were the most hilarious.
I watch these completely for the final voice at the end of AlphaZero.
You must have been very excited to watch it go fishing xd
yeah haha i remember the early bulletchess videos - jerry the fishing pole king =D
If you watch all of the games thus far played between AZ and Stockfish you will notice that at some point in the game Stockfish basically becomes "confused". There is a point in space in time where it really doesn't matter how deep Stockfish scans, that Alpha Zero has played a move that is not in "any" book or database. This is the beauty. This is why its an AI.
*Your beginners to chess master videos are great* ....*I have learnt a lot from them*...*thank you very much*....
👍🏼
12:35 the bishop is heavily restricting black's light-sq bishop. Note how tangled black's pieces are as well; trading would relieve some of that pressure.
They played Rhf1 because it keeps the d5 pawn supported, which may be helpful after e5.
Loving your content Jerry, keep up the good work!
What a great analysis!! Beautiful!!! Thank you very much.
You should write a chess book on alpha zero, Jerry.
Can Confirm I'd prefer he continue to make videos.
Great explaining man .....those lines helps a lot
Looking forward to seeing more of these games!
I live for the Alpha Zero voice scolding stockfish at the end.
@8:45 you try to explain the move pawn a4, throwing out possible moves like Ra4, and also explained why Bc3 isn't played because it could also block the rook after any Ra4 moves in future. But I think the most important is like you said, to restrict blacks pawns on the queenside, especially any quick b5/b4 moves, gaining time and space against any quick Bc3 moves. It also stops quick a5/a4 expansion ideas for good, so it's just a good space grabbing moves for white, and at the same time a queenside restricting move.
a4 is better than Bc3 because that only develops your piece but does not restrict black on queenside or anywhere. It attacks the knight but there is no rush to take the knight out. But there is a rush to hinder blacks queenside development/expansion before he hinders YOURS.
The choice between a4 and Bc3 was obvious with such logic/reasoning behind a4 compared to lack of logic behind Bc3. I'm sure engine calculation will also show a4 to be the better move in terms of hard calculation, it just grabs more space long term and is probably a "must play" move, else black will instantly play b5 and a5 himself if allowed IMO. Someone let me know if you know I'm wrong.. Don't mind if you have engine.
Great to see this series starting up again :D
What is disturbing is Alphazero is teaching itself . A.I. will rule the world, not just chess
Thanks, was waiting for this analysis. Looking forward on your insights on other released games. (I wonder why not release all games?)
@15:00 you say that you can't think of any reason for Rhf1 instead of Rdf1 but then mention some lines where white plays e5 dxe5, d6 pawn breaks, and if white had played Rdf1 the rook would no longer be protecting d5 from Bb7 playing Bxd5, so the e5 pawn break would no longer be an option if white plays the Rdf1 line..
Just throwing my own thoughts out there, same style as Jerry.. He inspired me. Haha. Hopefully it can help someone else, same as he helps me.. These games are very instructive indeed. Especially a4 and Rhf1.. Move order.. is everything!
G'day and happy chess to y'all... ; ]
love ur channel jerris keep up the good work
Me: loves to play chess with others
also Me: *Visible confusion watching these games*
Super video and narration. Easy sub. Thank you.
"That's... probably where my thoughts would end on this decision." Classic Jez.
Will you be doing more Leela videos? She is shooting up the ranks in TCEC in this season and will likely make it to premier division, and she has been placing #3 or #4 in CCCC.
I would LOVE to see the thousand-move games, where both computers use continuous draw threat to force moves to be made. ;) Please post one of those videos!
If you listen real hard you can hear AlphaZero saying "OK Boomer"
I'd like to see AlphaZero take on the latest version of Stockfish in a series of games with classic time controls using the same hardware for both machines. Everything they've done so far is still impressive, of course, but I would really like to witness a match like that to truly know how well it'd go.
Yes I think your title is appropriate -- many games of A0 v Stockfish do look like fishing expeditions. Dangle a pawn in front of him and trade it for a crushing position!
Why on earth did Stockfish play 12…P-KR3 when it knew it would do nothing to scare the Knight away? I would have thought something like 12…PKN3 would have been better?
Thanks Jerry.
12:50 I really like this example. Even though I've seen it 100s of times... I tend to get materialistic... whether its a pawn, or a rook. However, Something just clicked for me. And I can't say it's the same reason why alpha zero decided not to capture it... When I decided to re examie the position, I see White having 25 points of minor+ major pieces in play. Both Bishops, Rooks, and queen. I see black with 17. Rook, Black Bishop and queen. (26-17) 26 Points that are all very close to attacking the king.
If the White Bishop captures the rook it will shortly be good, but after the queen captures, Black is 2-3 moves away from getting the other rook in play + Black Bishop. So after Qxe8, Bishop f6, the rook on g8 can come in, or wait a move then come to the open file on f. This would then Be roughly 23 points for White ( 2 rooks, Queen, Bishop) vs The same 17 Points for Black (Queen, bishop, and Other rook)
Mind you I am still learning... I could be looking at this the wrong way, but it seems like it would lessen the attack on the King.
LEELA VS ALPHAZERO?
Or Az on CCC?
@@haromt3 The only way to truly know AZ's strength is if Deepmind Submited AZ to CCCC or TCEC. I would love to see AZ in CCCC/TCEC one day.
I think AlphaZero lost a handful of these games right?
Will you review some of those, and perhaps explain what you think went wrong?
From my observations from the released games, A0 lost mainly due to endgame blunders.
Alphazero didn't lose a single game
@@St3lf There are a couple lost games in the big new dump DeepMind's released.
View them here. deepmind.com/research/alphago/alphazero-resources/
@@St3lf AlphaZero lost 6 games, drew 839 games, and won 155. This gives AlphaZero a +52 Advantage over Stockfish 8.
@@Cscuile And how does that translate to elo? How much higher is Alpha's estimated elo?
It may not be any kind of crazy number for youtube but I'm very pleased to see a number like 167K views for this video.
How long did Google actually let AlphaZero train itself? And with how much computing power?
Thanks in advance!
Greetings from Switzerland.
ps: Looking forward to many more of those videos! During Carlsen vs Caruana I was actually wondering when they get released. And whether they will let them play again soon.
humans are playing more like computers while computers are playing more like humans
Yes! Thank you for showing us!
Maybe there some merit about waiting to Castle. Maybe the advantage that you get by not having to waste a move carries quite far into the game and that by the time you need to castle you’ve gained an advantage being one move ahead for several moves. It’s certainly something to consider.
New chess proverb? Castling in a calm opening is often aji-keshi?
I call it "flexible king"
I was looking for this game around RUclips. Thanks
14:55 Rdf1 vs Rhf1? If Rdf1 g3 later threatening g2 with fork?
#suggestion
TCEC 14, div 3, game 24.4. Leela vs Arasan.
This may not be very instructive from purely chess-y point of view, but may force to scratch one's head about what the hell is going on with those engines.
At first Leela easily wins a rook, but then shuflles for more than 50 moves struggling to convert (her evaluation was all that time just under treshold of 10).
Playing against AlphaZero would be so hopeless and miserable. That level of play is alien-like.
Alpha Zero was in a supercomputer. Stockfish was not. Unfair match, if you want my honest opinion.