Hi agadmator I think you are missing the point of the AlphaZero vs Stockfish series as well. AlphaZero is basically using the Google super computer and Stockfish doesn't run on that hardware; Stockfish was basically running on what would be my laptop. If you wanna have a match that's comparable you have to have Stockfish running on a super computer as well. AlphaZero would not have been able to reach the advantageous positions here in many cases if Stockfish was running on a super-computer.
Nowadays chess is much more solid and people believe that many of Tals sacrifices wouldn’t work. Alpha Zero proving that sacrifices still have potential is very interesting.
I always like to imagine Stockfish as a demigod of chess who laughs at peasant humans for not playing machine moves to win and there comes literal god Alpha Zero smirking and obliterates Stockfish with a human move.
I agree on what others say "show us a game where Stockfish wins", but I couldnt agree more when you said that it's not about who won the match, but how creative both players play their game and refute each other's plan. LOVE U AGAD!!
The one that people want to see is the one where Stockfish kept sacrificing pieces and won. It was pretty creative and entertaining. Here's Suren showing the game: ruclips.net/video/7JgR4iPeyxQ/видео.html
@John Doe Nope there are two different types of sacs and I didn't invent the idea. btw Just because a sac can't be determined 100% to be sound does not mean that it is a blunder.
@@MineCraftrules17 It is amazing that alpha zero got so good just by playing - don't know - millions of games against irself. I did not get better by playing against myself, the only thing was - i won nearly all the games exept some draws...
@@V8SupersQirreL I would assume that Alpha Zero doesn't really play against itself. At least for Deepmind's Alpha Star (the Starcraft 2 "AI") it's the case that there exist different versions of the "AI", called agents, playing against each other. Also the learning process doesn't really resemble human behavior. The engine starts playing by just knowing the rules of chess and then two randomized versions (regarding the decisions) start playing against each other without being biased by knowing any strategies.
@@tehjokur7041 I guess, i don't really understand what you mean - so you must be right! My english is not the best, what i told was the version i heard, i believed, and it was kind of realistic. Btw - my computer-knowledge is as good as my english!
These games were from early 2018. By now, analyses by SF9 and SF10 have discarded quite a few of SF8's choices. Looks like the programmers have made SF wiser too. AZ will still crush the newer versions, but not by this margin.
Stockfish is reigning world champion. AZ refuses to play Stockfish on even terms(like TCEC). Stockfish crushes everyone(including Leela) they put in front of it. Connect the dots folks.
The creator of Stock fish is a proud sponsor of Woke and black lives matter and does not appreciate your racist inspired comment. White men have gone first just too many times in chess and that's going to stop.
Honor of this e5-sacrifice, followed by f5, belongs to late Kaarle Ojanen. It is even named after him; every chessplayer in Finland knows the thing called "Ojasen oivallus" (=Ojanen's epiphany). This idea has been played in some number of occasions, but he was the one who found it first.
You are speaking of an idea a strategical motif. No one can say who saw this idea first it may have been first played in 1485. (Unless you mean in that exact position?)
You should watch videos from IMs or GMs. These shallow vids by this 2000 guy who barely speaks English don't add much to chess. He doesn't understand the games he shows on any level
@@paranchoy6077 I don't only watch the videos to get mind blown by the interpretation of the chess moves. I watch him because he is funny, wholehearted and entertaining. And I think his subscriber count speaks for itself. If you look for a deep analysis of these games you might be right, but most people don't. We just want to be entertained. There's nothing wrong with that. :)
@@bustarogers9990 it was a joke and I can't tell if you are upset or joking also. sheesh I should probably add a, (joke) to all my terrible joking comments to clarify
Watching a video overseas on holiday in China, so I had to buy data and get a VPN just to watch you, almost 1 min in when u uploaded, at 1am after a long flight and a long day, because your content is that great. Thanks and keep it up Adgamator! :) Also #suggestion apparently there was a game that Stockfish won against A0 where Stockfish was super aggressive and won brilliantly. Please:)
@@rahulmalhan2262 it's not illegal but most Google related things and things like Instagram are blocked. They have a country wide firewall that prevents this
@@rahulmalhan2262 There is a firewall but when you are a foreigner (and even for Chinese but they do not really need it and they face harder reprimands) it is easy to get a VPN and access to any website
1.) Go to 11:35 2.) Close your eyes 3.) Imagine Agadmator doing a impression of kermit the frog, doing a chess video 4.) Unpause the video 5.) ... Enjoy :DD
I closed my eyes and imagined Kermit analysing a chess game between Beaker and Animal. How long will it take before I can get that idea out of my head?
@@ClearReception - brilliant! Also (courtesy of Kermit's little nephew Robin - original by A A Milne): Half way up the squares is the square where I sit
The foresight of that rook to c7 move is brilliant. On the surface it seems like a harmless repositioning of the rook, but after a few moves and the exchange of the queens, it becomes a positional monster. It's beautiful
Great game. At first I wasn't sure if I wanted to watch a game between engines, but after seeing these moves and your brilliant explanation of the ideas behind them I'm looking forward to watching more!
Stockfish 8 on i5 given a couple of minutes to think: 7:18 Stockfish only looks deeply into 26.a3 but e5, f5 feature in most of the lines. 3 moves later, it takes half a minute before it begins to score Alpha Zero's line better. 12:20 Stockfish does find and prefer 35.Rc7 straight away.
These games have really been a treat. The primary themes seem to be stifling development, opening files and diagonals, and cutting pieces out of the game.
9:26 Yeah black's bishop is hanging but just pushing b3 defends. Stockfish seems to think both that continuation and bishop b3 are equal (equally bad that is, white has like +1.2 or something in both of them) just because black's position is worse here, i don't see any major reason why it has much to do with whether black pushes a3 or tries bishop b3. Wouldn't be surprised if I'm missing something though but stockfish also doesn't seem to find a huge problem with pushing b3 either in the following move.
Not true. The main 1000-game match where AlphaZero lost 6 games was played without opening books. Further matches were played with different kinds of opening configurations (human openings, TCEC openings and Stockfish with book versus AlphaZero with no book). In those matches, AlphaZero won by similar or slightly lower margins and also lost a few games.
@Time Warp this isn't really how an AI like alpha zero learns. The entire idea behind neural network based approaches to machine learning is that the resulting network will generalize onto the entire distribution of data, even when only trained on a small subset of the distribution. Alpha knows enough about strategy to understand how to play, even when it's never seen the exact state of the board before. What you're describing is actually way more similar to how a chess engine would learn. Basic engines work by just creating a type of decision tree and assigning a score to each branch based on some heuristic.
@Time Warp no not really. It's slightly more complicated than what I am describing, but essentially the purpose of neural network machine learning is to train the network to recognize complex patterns in some data set (all of the games that alpha zero played while training) while at the same time making sure that the algorithm can still make relevant predictions in more generalized scenarios. If AZ could only succeed on states it has already seen, it would not be successful in any chess game against a human or stockfish (this is what would be called overfitting: the AI is only good on the specific data it has already seen). It's easy to understand this intuitively, because most chess games reach a point where that exact state has never before been seen, yet AZ still will win the vast majority of those states against any human.
The 6 games alpha lost were because they run the current version of stockfish on the same hardware as alpha. After losing 6 games google thought it would be bad PR if alpha loses even more games so they decided to use an old version of stockfish and let it run on a cheap pc while alpha runs on a supercomputer, and voila alpha won all the other games.. alpha is practically the Tal Baron of computerchess, only winning if cheating
@@PeterPan-uf9or why do I choose tal baron when there are thousands of cheaters and dozens of GMs cought cheating. Also I would rather call a cheater to someone who cheats in real life tournament than in some online match
11:58 the Knight can still go out of harms way with g5. Black's rook still moves to protect the Knight. And then white Knight to e6. I don't understand the genius behind white rook from a7 to c7. Can anyone explain?
20 moves after Rc7, When the pawn comes to d7, the rook threatens Rc8 winning the black rook. Basically Rc7 gains a key tempo for free It also hits c5 but it's not like Alpha cares about grabbing pawns :)
Actually no. Alpha0 plays as if they made a copy of Garri Kasparow and gave Garri unlimited calculation power. Tal was positionally often unsound but his opponents couldn't find counterplay at the board. Kasparow was the most dynamic player and the most crushing in complex dynamic positions, while actually being positionally supersound and often demonstrating an amazing understanding of those sharp lines any other player tried to avoid. That's why it was so surprising and kind of silly that he lost against Kramnik in 2000. Instead of playing to win the match no matter what, he started a theoretical dispute on the board over the Berlin defense ...as if this was an endless match versus Karpow from the 80s. He wanted to be right yet again and break the Berlin defense by going with his head through the wall. Kramnik said he hoped for this psychological momentum vs. Kasparow because it was the only way to win. And he pulled it off, despite Kasparow being the stronger player and remaining the stronger player after 2000.
14:42 Why dosn¨t knight captture on b2? I mean, by first checking on C4 then capture on b2? After white pawn goes to d7, rook can block with d8, and after rook c8 the king can go to e7 protecting the rook and threatening the pawn?
if u go knight c4 check white would go f3 or f4 and if u take b2 white would take c5 and attack the pawn on b3 and u cant defend it with knight or check the white king u can only protect it with rook b8 or lose it too after Rb8 white pawn will go d7 and u cant stop it with king if u go Ke7 u will get Rc8 and white will get ur rook or get a queen. u can block the e pawn with your rook but u would lose all of your queenside pawns. which is the only counterplay you have.
Rc7 isn't a move you find by calculating the position. it is just impossible to calculate such a move. I you found it by luck let me ask you something: would you have played it in a real important match like in the world chess championship match over the board?
Please do all the games!! AlphaZero is amazing, this is a new paradigm for chess. Question - is A0 getting better with every game? Or is it at its peak?
actually AI and nanotechnology could be used to manipulate our biology on a molecular level, unlocking potential that could far outstrip anything in the world of machinery. humans building machines superior to humans, so that machines can build humans superior to machines.
@@uberneanderthal probably not, if you'd ever worked with it you'd know those things are extremly unreliable, they woud bess you up more times than not,
When you mentioned some unusual move, I immediately considered e5. That's a very common thematic pawn sac against the benoni and the like. f5 alone leaves black a gaping hole for counter play, e5 forces black to plug the hole and usually with white's e4 control, black is going to have to give back the pawn to create counter-play (like Stockfish eventually did)
Watching alphazero suffocate his opponent whilst expanding positioning is genius. When will Stockfish realise if Alphazero feeds him a pawn it's a trap... Many thanks for your time in analysing and presenting these amazing games...
This is literally paradigm shifting. Chess will witness a paradigm shift in thought and action from anyone who is growing up and learning the game right now as a child. We'll only see the full impact of these revelations embedded into human intuition and trust with the next generation of players.
A big part of research is getting financial and public support to keep doing what they do, and hopefully sell it. Lot of potential for this technology, Chess and Go are just one of the many areas that have benefited from it and the perfect place to showcase it.
It would be interesting if you could display stockfish's evaluation of the position after each move. I'm curious to see whether stockfish ever thinks it's better and just how fast it loses its advantage.
From what I have heard all victories of Stockfish come from Alpha being forced into a human opening so Stockfishes victories probably arent that much through great plays but because those openings are bad for Alphas Playstyle.
@@evgiz0r Yeah Alpha Zero is not invincible in chess (yet) like it is in go where the only AI that can hold a candle are other AlphaGo AIs and where humans are just helpless against AlphaGo exactly because Go is so complicated that any not neural network AI has no chance and humans have an immnse disadvantage in processing power. Chess is much more close to being solved which makes conventional algorithms really strong, but again those games where A0 loses probably arent too eciting because it is probably more Alpha playing bad than Stockfish playing really good because these openings dont have those weaknesses with other engines. This seems to be a special weakness for A0 hich is great for DeepMind because that means that A0 can still improve by a lot in chess.
@@Beertraps About half of the loses I've seen from AlphaZero came from it not wanting to accept a three-fold draw so it pushed on even with it then being in a worse position
@@carrottoponcrak Interesting. Is that from the current AlphaZero or from the old ones? I havent read the papers myself, but I have seen numerous comments stating that all losses came from the games where it had to do human openings. Maybe the evaluation/policy network is really inaccurate in those bord states because they are in the developed early game and therefor still quite complex and the neural network has almost zero experience with them because it discarded them quite quickly. This could mean that A0 thinks a board state to be much better than it really is which makes it logical to push for a win. Are those pushes still very early? Because a push that puts you into a worse position (in terms of material for example) can increase winning percentage and in difference to stockfishes programming A0s evaluation network is only predicting winning probabilities of a board state. maybe what I am speculating here is compltely false. No matter what the reason is I highly expect DeepMind to find that out and try to improve on that. There was a similar problem in AlphaGo Lee where it misjudged boardstates for many turns and therefore made really bad moves.
I agree with the "pleasant to the human eye" part. I watched some of the engine championship matches and many were not pleasant. Like 180 moves with 30 consecutive moves just constantly adjusting 2 or 3 different pieces slightly without 3 repeats. Those games were not pleasant to the human eye.
14:41 why not knight c4 check by Stockfish? King D4 most likely then knight takes pawn, then just move the knight out of the way for stockfish pawn to promote to queen and defend it with rook b8
I liked the video, but I have to nitpick: you never really explained why Rc7 was such a good move -- even with the white rook on a7, ..Re8 could be met by Ng5, Rd8, Ne6, Re8, Nf4 etc. -- Im not sure if the point was strictly so that after Nf4 Ne5 d7 Rd8 can be met with Rc8 (which obviously if the rook was still on a7, it wouldn't be able to play Rc8) or if there was some other move/line that AlphaZero was actually parrying? It just wasn't very clear what exactly the point of Rc7 was from the video since you made it seem like Ng5 wouldn't be possible with the white rook on a7, but clearly it was still possible. Also, given how forcing all of these moves were, I could actually see any number of good grandmasters finding Rc7 pretty easily -- just my two cents.
I think the Rc7 move prevents the black knight on d7 from ever moving because it's kinda stuck on guarding the c5 pawn, as well as the black rook it's stuck on guarding the black knight, and because of those the white knight can freely capture the black pawn on the king's side.
It seems as though Stockfish 8 repeatedly underestimates the power of pawn sacrifices to gain positional advantage. Stockfish 8 always seems to play a balanced game but cannot handle Alpha Zero’s masterful ability to obtain positional dominance. Simply brilliant!
Esse Alphazero parece uma junção de dos melhores jogadores da história: sacrifícios de Tal, precisão de Fischer, aperta o adversário como Karpov, tem a frieza de Carlsen e faz tudo pela atividade das peças como Kasparov. Na minha opinião o melhor jogador de xadrez que pode existir! Cadê os Br?
To be honest I haven't enjoyed chess as much as this in quite a long time after watching the A0 matches. Im really looking forward to the day we can watch live matches between two creative AI`s with analysis from human grandmasters.Their passion for the game actually makes it even more enjoyable to watch.
Hey agadmator, it'll be awesome if you make a video on Chess theories, and how some needs to be broken (like alpha does) and how with theory one can develop the openings without even knowing openings much. Thank you.
Did they play these positions from opposite sides at some point? Alpha with the King’s Indian and Stockfish with white? Would be interesting to see what Alpha recommends in the defensive moves leading up to the Tal Pawn sac.
@13:09 Why wouldn't the same idea work if the rook was on a7? It seems to me it was just a positional advantage, defending the rook with the passed pawn and attacking black's pawn. But regarding the knight moving to g4 nothing seems to be different.
i run it on stockfish 15 and i was amazed! I read everywhere the new stockfish would destroy the old alphazero, and with the advancing of computing and programming i thought it makes sense. Actually even the most recent stockfish is still blind about that pawn move. By the way would someone explain me how rook to c2 changed anything? in the previous position with the rook on c7 the knight apparently could still simply jump to f4 but instead the rooks would just take the knights according to Agad. Why is that?
If you were able to see that the light is shining on his hair congratulations you are a great moth
Bröther... pass me the lämp
😂😂😂😂
You forget. IF he speaks as a hypothesis, THEN he must finish as a hypothesis
Hi agadmator I think you are missing the point of the AlphaZero vs Stockfish series as well. AlphaZero is basically using the Google super computer and Stockfish doesn't run on that hardware; Stockfish was basically running on what would be my laptop. If you wanna have a match that's comparable you have to have Stockfish running on a super computer as well.
AlphaZero would not have been able to reach the advantageous positions here in many cases if Stockfish was running on a super-computer.
@@TheTinnin boy you are so wrong...
Why isn't the fishy guy winning? Rating 8 might not be great, but atleast it's more than Alpha's zero.
pu tum ptzz.
Brilliant comment. Best laugh I've had all week (including last week!)
Haha
This took me way too long to get.
Elder god tier joke.
tal: *sacs queen and 3 rooks then wins*
everyone: its just normal tal play
computer: *sacs pawn*
everyone: *faints*
Yes.... That's the actuality.....
I'd like to see any human sac a pawn to stockfish and win
@@slamalamadingdangdongdiggy5268 yup alpha sac 2 pawns in position where stock fish had seemingly invincible pawn push on queen side and alpha wins
Nowadays chess is much more solid and people believe that many of Tals sacrifices wouldn’t work. Alpha Zero proving that sacrifices still have potential is very interesting.
@@slamalamadingdangdongdiggy5268 🤣🤣🤣🤣🤣🤣
I always like to imagine Stockfish as a demigod of chess who laughs at peasant humans for not playing machine moves to win and there comes literal god Alpha Zero smirking and obliterates Stockfish with a human move.
who is Lc0
Well said
The prophet has spoken. That'd make a sick film.
@@cadespaulding3837 Lc0 is leela chess zero
Lc0/Leela is actually open source and based on alpha
Just the hearing of the word ''Tal'' is enough to brighten your day
i think you meant to say word but the sentence makes sense nonetheless
eh, Tal runes aren't that great
I agree on what others say "show us a game where Stockfish wins", but I couldnt agree more when you said that it's not about who won the match, but how creative both players play their game and refute each other's plan. LOVE U AGAD!!
The one that people want to see is the one where Stockfish kept sacrificing pieces and won. It was pretty creative and entertaining. Here's Suren showing the game: ruclips.net/video/7JgR4iPeyxQ/видео.html
@@DezDav4 +
@@DezDav4 The only way SF sacs pieces is if it is not a true sac. ie. if it can calculate a mate or later regaining material achieving a won position.
@John Doe Nope there are two different types of sacs and I didn't invent the idea. btw Just because a sac can't be determined 100% to be sound does not mean that it is a blunder.
@John Doe Rudolf Spielmann for one.
Alpha
He protec
He attac
But most importantly
He pawn sac
I chuckled!
Cringe
Can someone please tell me where the hell that "he protec , he attac , but" line came from????. It's every freaking where.
@@bustarogers9990 It's a meme.
Patrick Salhany underrated comment 😂
With all the imbalances that Alpha Zero makes, I would be really interested to see how Alpha Zero would play against itself.
I mean.... That is how it learns
@@MineCraftrules17 It is amazing that alpha zero got so good just by playing - don't know - millions of games against irself. I did not get better by playing against myself, the only thing was - i won nearly all the games exept some draws...
@@V8SupersQirreL I would assume that Alpha Zero doesn't really play against itself. At least for Deepmind's Alpha Star (the Starcraft 2 "AI") it's the case that there exist different versions of the "AI", called agents, playing against each other. Also the learning process doesn't really resemble human behavior. The engine starts playing by just knowing the rules of chess and then two randomized versions (regarding the decisions) start playing against each other without being biased by knowing any strategies.
@@tehjokur7041 I guess, i don't really understand what you mean - so you must be right! My english is not the best, what i told was the version i heard, i believed, and it was kind of realistic. Btw - my computer-knowledge is as good as my english!
@@V8SupersQirreL I found it was very difficult to play against myself in chess, so long as I was making an honest attempt with both pieces.
Only dislike is from stockfish 8
Alpha deleted that dislike
Right... Only a moron would play against SF 8. SF 10 will crush you in half the moves
These games were from early 2018. By now, analyses by SF9 and SF10 have discarded quite a few of SF8's choices.
Looks like the programmers have made SF wiser too.
AZ will still crush the newer versions, but not by this margin.
Stockfish is reigning world champion. AZ refuses to play Stockfish on even terms(like TCEC). Stockfish crushes everyone(including Leela) they put in front of it. Connect the dots folks.
@Brad Heilman That's the anarchy symbol, bro, not atheism. If you're going to be an asshole, at least don't be a stupid asshole.
TALpha
And I thought I was being clever Tal-pha Zero... :P
@@John_II TALpha requieres no zero because it has Tal's left hemisphere which adds so much elo not to permit a numerical quantification in any field 😂
@@gioser3021 science, bitch!
Alpha ZeTal
I think that this all is propaganda. In do not believe that Stockfisch is so weak against alpha z.
8:56 -
Stockfish’s most powerful piece on h6 =
A queen in prison
Riker’s Island Correctional Facility = a prison in Queens
HAHAHA. Yup. And there are sometimes queens in that prison for rolling customers...
J J good one
I legit live 2 blocks away from rikers island in queens NY.. Where can i find this queen ;)
Only problem is that Rikers is in the Bronx, hahaha...
"Playing basketball in Pelican Bay!!!"
Is stockfish never on white?
All the matches were 50/50 proportion white and black.
Lol weird seeing a channel like this on a channel like this
I think stockfish managed to draw most (all?) of the games it played as white, so perhaps it's not as interesting to review :)
The creator of Stock fish is a proud sponsor of Woke and black lives matter and does not appreciate your racist inspired comment.
White men have gone first just too many times in chess and that's going to stop.
@@ericmol5698 the right can't meme
Honor of this e5-sacrifice, followed by f5, belongs to late Kaarle Ojanen. It is even named after him; every chessplayer in Finland knows the thing called "Ojasen oivallus" (=Ojanen's epiphany). This idea has been played in some number of occasions, but he was the one who found it first.
You are speaking of an idea a strategical motif. No one can say who saw this idea first it may have been first played in 1485. (Unless you mean in that exact position?)
You know, the thing!
Holy shit I was scared you wouldnt upload today. I need this daily content hahaha
(Nervous laughter)*
jea basicly, its like drugs for me 😂
ITS EVERYDAY BROOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
You should watch videos from IMs or GMs. These shallow vids by this 2000 guy who barely speaks English don't add much to chess. He doesn't understand the games he shows on any level
@@paranchoy6077 I don't only watch the videos to get mind blown by the interpretation of the chess moves. I watch him because he is funny, wholehearted and entertaining. And I think his subscriber count speaks for itself.
If you look for a deep analysis of these games you might be right, but most people don't. We just want to be entertained. There's nothing wrong with that. :)
Tal's spirit is in alpha.
The anime is coming soon
@@bustarogers9990 it was a joke and I can't tell if you are upset or joking also. sheesh I should probably add a, (joke) to all my terrible joking comments to clarify
It's so not so much Tal, but clearly Kasparow.
Tal understood everything
felt *
is dead*
@@tristanperrow8518 Is English your first language? The OP used the past tense.
@@dannygjk lol
Watching a video overseas on holiday in China, so I had to buy data and get a VPN just to watch you, almost 1 min in when u uploaded, at 1am after a long flight and a long day, because your content is that great. Thanks and keep it up Adgamator! :)
Also #suggestion apparently there was a game that Stockfish won against A0 where Stockfish was super aggressive and won brilliantly. Please:)
He mentioned the game . It’s because they set up an opening book that wasn’t real
That‘s some dedication there! Well done!
Isn't using RUclips illegal there.. Or it's allowed for foreigners??
@@rahulmalhan2262 it's not illegal but most Google related things and things like Instagram are blocked. They have a country wide firewall that prevents this
@@rahulmalhan2262 There is a firewall but when you are a foreigner (and even for Chinese but they do not really need it and they face harder reprimands) it is easy to get a VPN and access to any website
12:08 "And it will be a very nice draw." Perfect 😁
Exactly a Tal move; we've seen him make moves just like that. Thank you for finding that move in the huge release of Alpha Zero games.
Imagine in 30 years, GMs who grew up learning chess by Alpha Zero
Alpha0 plays like Kasparow and Iwantschuk.
I found the move Rook c7, wasn`t so hard.....
after i watched this vid 3 times.
LOL I believed in humanity for 1 second
I have to say i was away from xhess for about 6 months and in that time it seems your analysis which was excellent has gotten even better
1.) Go to 11:35
2.) Close your eyes
3.) Imagine Agadmator doing a impression of kermit the frog, doing a chess video
4.) Unpause the video
5.) ... Enjoy
:DD
I closed my eyes and imagined Kermit analysing a chess game between Beaker and Animal.
How long will it take before I can get that idea out of my head?
I hate you lol
It's not easy being blacks queen.
hhhhhhhhhh 😂😂😂😂😂😂😝😝😝
@@ClearReception - brilliant!
Also (courtesy of Kermit's little nephew Robin - original by A A Milne):
Half way up the squares is the square where I sit
The foresight of that rook to c7 move is brilliant. On the surface it seems like a harmless repositioning of the rook, but after a few moves and the exchange of the queens, it becomes a positional monster. It's beautiful
yay Finally we have the Evan's gambit on board !
Great game. At first I wasn't sure if I wanted to watch a game between engines, but after seeing these moves and your brilliant explanation of the ideas behind them I'm looking forward to watching more!
14:29 you can even blunder, blunder , mate in 1❓❓#️⃣
*Always my style to get checkmate* 😎😎😎
Ahhh it's so satisfying to hear in that position stockfish resigned the game
Commenting about a 20 minute long video after just 10 minutes. Watching level: AlphaZero
USE 2X SPEED
@@elevengiant thats something Stockfish would do
Stockfish 8 on i5 given a couple of minutes to think:
7:18 Stockfish only looks deeply into 26.a3 but e5, f5 feature in most of the lines. 3 moves later, it takes half a minute before it begins to score Alpha Zero's line better.
12:20 Stockfish does find and prefer 35.Rc7 straight away.
When you say "Stockfish immadiately captures" i imagine a fish playing against a brain, and the fish just blitzes out a move
Love your channel! Am getting better when I can’t play any longer I just watch:) thanks!!
Beautiful maneuver in the centre... Loving Alpha's tactics against stockfish... you are right, future is bright for chess 👍
These games have really been a treat. The primary themes seem to be stifling development, opening files and diagonals, and cutting pieces out of the game.
In some videos, like in this one, the only info in the info corner is stating that the respective corner is the info corner
wow! that is crazy!!! mind blown, think i replayed that c7 move like 10 times. Beautiful game, one of my favorites thus far.
the rook c7 move was preparing for a position 13 moves ahead. against a ~3400 elo opponent. incredible.
9:26 Yeah black's bishop is hanging but just pushing b3 defends. Stockfish seems to think both that continuation and bishop b3 are equal (equally bad that is, white has like +1.2 or something in both of them) just because black's position is worse here, i don't see any major reason why it has much to do with whether black pushes a3 or tries bishop b3. Wouldn't be surprised if I'm missing something though but stockfish also doesn't seem to find a huge problem with pushing b3 either in the following move.
I remember hearing that the 6 games that Alpha Zero lost, it's because humans forced it to play openings that humans would play.
Not true. The main 1000-game match where AlphaZero lost 6 games was played without opening books. Further matches were played with different kinds of opening configurations (human openings, TCEC openings and Stockfish with book versus AlphaZero with no book). In those matches, AlphaZero won by similar or slightly lower margins and also lost a few games.
@Time Warp this isn't really how an AI like alpha zero learns. The entire idea behind neural network based approaches to machine learning is that the resulting network will generalize onto the entire distribution of data, even when only trained on a small subset of the distribution. Alpha knows enough about strategy to understand how to play, even when it's never seen the exact state of the board before.
What you're describing is actually way more similar to how a chess engine would learn. Basic engines work by just creating a type of decision tree and assigning a score to each branch based on some heuristic.
@Time Warp no not really. It's slightly more complicated than what I am describing, but essentially the purpose of neural network machine learning is to train the network to recognize complex patterns in some data set (all of the games that alpha zero played while training) while at the same time making sure that the algorithm can still make relevant predictions in more generalized scenarios. If AZ could only succeed on states it has already seen, it would not be successful in any chess game against a human or stockfish (this is what would be called overfitting: the AI is only good on the specific data it has already seen). It's easy to understand this intuitively, because most chess games reach a point where that exact state has never before been seen, yet AZ still will win the vast majority of those states against any human.
The 6 games alpha lost were because they run the current version of stockfish on the same hardware as alpha. After losing 6 games google thought it would be bad PR if alpha loses even more games so they decided to use an old version of stockfish and let it run on a cheap pc while alpha runs on a supercomputer, and voila alpha won all the other games.. alpha is practically the Tal Baron of computerchess, only winning if cheating
@@PeterPan-uf9or why do I choose tal baron when there are thousands of cheaters and dozens of GMs cought cheating. Also I would rather call a cheater to someone who cheats in real life tournament than in some online match
11:58 the Knight can still go out of harms way with g5. Black's rook still moves to protect the Knight. And then white Knight to e6.
I don't understand the genius behind white rook from a7 to c7.
Can anyone explain?
20 moves after Rc7, When the pawn comes to d7, the rook threatens Rc8 winning the black rook. Basically Rc7 gains a key tempo for free
It also hits c5 but it's not like Alpha cares about grabbing pawns :)
Its a waiting move.
They installed tal spirit into alpha zero.
Actually no. Alpha0 plays as if they made a copy of Garri Kasparow and gave Garri unlimited calculation power.
Tal was positionally often unsound but his opponents couldn't find counterplay at the board. Kasparow was the most dynamic player and the most crushing in complex dynamic positions, while actually being positionally supersound and often demonstrating an amazing understanding of those sharp lines any other player tried to avoid.
That's why it was so surprising and kind of silly that he lost against Kramnik in 2000. Instead of playing to win the match no matter what, he started a theoretical dispute on the board over the Berlin defense ...as if this was an endless match versus Karpow from the 80s. He wanted to be right yet again and break the Berlin defense by going with his head through the wall. Kramnik said he hoped for this psychological momentum vs. Kasparow because it was the only way to win. And he pulled it off, despite Kasparow being the stronger player and remaining the stronger player after 2000.
14:42 Why dosn¨t knight captture on b2?
I mean, by first checking on C4 then capture on b2?
After white pawn goes to d7, rook can block with d8, and after rook c8 the king can go to e7 protecting the rook and threatening the pawn?
if u go knight c4 check white would go f3 or f4 and if u take b2 white would take c5 and attack the pawn on b3 and u cant defend it with knight or check the white king u can only protect it with rook b8 or lose it too after Rb8 white pawn will go d7 and u cant stop it with king if u go Ke7 u will get Rc8 and white will get ur rook or get a queen. u can block the e pawn with your rook but u would lose all of your queenside pawns. which is the only counterplay you have.
white can play nf6 threatening the rook and a check luring the king away from the passed pawn
I found 4 moves in a row by Alpha, including Rc7
It made me happy
ur a god
Don't lie, rook to c7 is not possible to find for human
@@ritikbhimkar4122 It is if you’re lucky
@@sungod9797 fax bro, i mean u can find it and dhope u are right
Rc7 isn't a move you find by calculating the position. it is just impossible to calculate such a move. I you found it by luck let me ask you something: would you have played it in a real important match like in the world chess championship match over the board?
Alpha is such a powerful AI it defeated Casablanca and Carlsen in a poll when it wasn’t even an option.
Please do all the games!! AlphaZero is amazing, this is a new paradigm for chess.
Question - is A0 getting better with every game? Or is it at its peak?
really great that we can see these games. Thanks for your commentary /analysis, very much appreciated.
-> Chess has a bright future
-> Humanity does not
The first step is amazon fire :(
this aged well
actually AI and nanotechnology could be used to manipulate our biology on a molecular level, unlocking potential that could far outstrip anything in the world of machinery.
humans building machines superior to humans, so that machines can build humans superior to machines.
@@uberneanderthal :00000
@@uberneanderthal probably not, if you'd ever worked with it you'd know those things are extremly unreliable, they woud bess you up more times than not,
When you mentioned some unusual move, I immediately considered e5. That's a very common thematic pawn sac against the benoni and the like. f5 alone leaves black a gaping hole for counter play, e5 forces black to plug the hole and usually with white's e4 control, black is going to have to give back the pawn to create counter-play (like Stockfish eventually did)
It truly says something when a chess engines take years of learning to make a Tal move for a win 😁
AlphaZero always gets stockfish in the most passive defensive positions locking in his own pieces. that queen prison is beautiful
Watching alphazero suffocate his opponent whilst expanding positioning is genius. When will Stockfish realise if Alphazero feeds him a pawn it's a trap...
Many thanks for your time in analysing and presenting these amazing games...
This is literally paradigm shifting. Chess will witness a paradigm shift in thought and action from anyone who is growing up and learning the game right now as a child. We'll only see the full impact of these revelations embedded into human intuition and trust with the next generation of players.
What's up with Google providing all the content creaters with alpha zero games?
Whats the idea here?
Marketing, d'oh. They probably want to sell DeepMind's thingamabob for some other purpose, such as machine translation or phone calls.
Publicity for sure. But I'm sure they're also quite proud of their work and what it's producing. I'd be desperate to share them if I worked there
A big part of research is getting financial and public support to keep doing what they do, and hopefully sell it. Lot of potential for this technology, Chess and Go are just one of the many areas that have benefited from it and the perfect place to showcase it.
It would be interesting if you could display stockfish's evaluation of the position after each move. I'm curious to see whether stockfish ever thinks it's better and just how fast it loses its advantage.
So, there are no interesting victories of Stockfish. Are there interesting ("creative") moves from Stockfish?
From what I have heard all victories of Stockfish come from Alpha being forced into a human opening so Stockfishes victories probably arent that much through great plays but because those openings are bad for Alphas Playstyle.
@@Beertraps it still says something about not being a complete monster :)
@@evgiz0r Yeah Alpha Zero is not invincible in chess (yet) like it is in go where the only AI that can hold a candle are other AlphaGo AIs and where humans are just helpless against AlphaGo exactly because Go is so complicated that any not neural network AI has no chance and humans have an immnse disadvantage in processing power. Chess is much more close to being solved which makes conventional algorithms really strong, but again those games where A0 loses probably arent too eciting because it is probably more Alpha playing bad than Stockfish playing really good because these openings dont have those weaknesses with other engines. This seems to be a special weakness for A0 hich is great for DeepMind because that means that A0 can still improve by a lot in chess.
@@Beertraps About half of the loses I've seen from AlphaZero came from it not wanting to accept a three-fold draw so it pushed on even with it then being in a worse position
@@carrottoponcrak Interesting. Is that from the current AlphaZero or from the old ones? I havent read the papers myself, but I have seen numerous comments stating that all losses came from the games where it had to do human openings. Maybe the evaluation/policy network is really inaccurate in those bord states because they are in the developed early game and therefor still quite complex and the neural network has almost zero experience with them because it discarded them quite quickly. This could mean that A0 thinks a board state to be much better than it really is which makes it logical to push for a win. Are those pushes still very early? Because a push that puts you into a worse position (in terms of material for example) can increase winning percentage and in difference to stockfishes programming A0s evaluation network is only predicting winning probabilities of a board state.
maybe what I am speculating here is compltely false. No matter what the reason is I highly expect DeepMind to find that out and try to improve on that. There was a similar problem in AlphaGo Lee where it misjudged boardstates for many turns and therefore made really bad moves.
Stockfish: I calculate to win.
Alphazero: Prepare the amada. We shall use the old ways.
"captures, captures"
I agree with the "pleasant to the human eye" part. I watched some of the engine championship matches and many were not pleasant. Like 180 moves with 30 consecutive moves just constantly adjusting 2 or 3 different pieces slightly without 3 repeats. Those games were not pleasant to the human eye.
I want to see alphazero vs alphazero.
Hopefully the world doesnt end by there game
your wish came true
14:04 "No longer the knight is pinned" - Yedi master allways right he is!
Bring back smiley pillow please.
I'm hoping Medo killed it.
You can buy 10,000 smiley pillow and put them over your desktop and monitor
glad to see you feeling better agadmator!!
Good
Better
Best
and then
Antonio does the rest!!!!
I really like these A0 vs Stockfish games. What I'd love to see is a A0 vs A0 game
@12:20 Instead of asking us to pause the video, he explains like for 20 seconds that he doesn't want to waste time?! 😂
"Stockfish doesn't actually resign." AlphaZero- "hold my beer."
The stockfish win against the french defense was spectacular come on ...
It was a janky french lol. White had too much
14:41 why not knight c4 check by Stockfish? King D4 most likely then knight takes pawn, then just move the knight out of the way for stockfish pawn to promote to queen and defend it with rook b8
I liked the video, but I have to nitpick: you never really explained why Rc7 was such a good move -- even with the white rook on a7, ..Re8 could be met by Ng5, Rd8, Ne6, Re8, Nf4 etc. -- Im not sure if the point was strictly so that after Nf4 Ne5 d7 Rd8 can be met with Rc8 (which obviously if the rook was still on a7, it wouldn't be able to play Rc8) or if there was some other move/line that AlphaZero was actually parrying? It just wasn't very clear what exactly the point of Rc7 was from the video since you made it seem like Ng5 wouldn't be possible with the white rook on a7, but clearly it was still possible. Also, given how forcing all of these moves were, I could actually see any number of good grandmasters finding Rc7 pretty easily -- just my two cents.
Yeah I too want to know:)
I think bc of Nxc5
Same, please explain :)
13:44... although he didn't stress the point at this juncture, but emphasized it at 13:57
I think the Rc7 move prevents the black knight on d7 from ever moving because it's kinda stuck on guarding the c5 pawn, as well as the black rook it's stuck on guarding the black knight, and because of those the white knight can freely capture the black pawn on the king's side.
It seems as though Stockfish 8 repeatedly underestimates the power of pawn sacrifices to gain positional advantage. Stockfish 8 always seems to play a balanced game but cannot handle Alpha Zero’s masterful ability to obtain positional dominance. Simply brilliant!
Guys does someone know where to watch the london chess classic?
Commenting just to also get the answer
Chess24
chess24.com/en/watch/live-tournaments/london-chess-classic-2018/1/1/1
prolly in london
I like how Alpha Zero looks for ways to win rather than ways to draw, unlike our human competitors that try to draw every game.
HELLO EVERYONE!
HELLO EVERYONE! HELLO FIRHAN STYLE!
Hello hello😊
Kings Indian Defense Saemisch variation is what i love to play.. it gets crazy like this... Tal, Fischer, Morphy would all be proud
so alpha zero also watches agadmator huh??
1:32 It has Zugzwang
A German loves this
Last!
Esse Alphazero parece uma junção de dos melhores jogadores da história: sacrifícios de Tal, precisão de Fischer, aperta o adversário como Karpov, tem a frieza de Carlsen e faz tudo pela atividade das peças como Kasparov. Na minha opinião o melhor jogador de xadrez que pode existir!
Cadê os Br?
I’ve been waiting for the words Alphazero and Tal
You seek brilliant jewels & this one has a golden setting of imaginative richness & extravaganza!! Thanks!!
To be honest I haven't enjoyed chess as much as this in quite a long time after watching the A0 matches. Im really looking forward to the day we can watch live matches between two creative AI`s with analysis from human grandmasters.Their passion for the game actually makes it even more enjoyable to watch.
It's really nice to see alpha zero!! It's just like the title says Chess is not dead because of theory. Bobby Fischer would smile.
You do a really great job reviewing the games. I'm glad I found you.
Hey agadmator, it'll be awesome if you make a video on Chess theories, and how some needs to be broken (like alpha does) and how with theory one can develop the openings without even knowing openings much. Thank you.
Incredible! So that lateral rook move was just so the passed e pawn can defend it after all those exchanges.
It seems so simple, the way if finds counter play and opportunities.
10:50 what's wrong with nb6. it guards the pawn and that knight wasn't doing anything anyway. it can also go to nc4 in the future
Bishop takes took on e8 in response, if I read your timecode correctly
The only way I found c7 was if in a mouse slip. Don't worry about doubting us
Lol, watching this again a year later and I almost commented the same thing. I guess I'm one dimensional.
Did they play these positions from opposite sides at some point? Alpha with the King’s Indian and Stockfish with white? Would be interesting to see what Alpha recommends in the defensive moves leading up to the Tal Pawn sac.
Alpha zero vs alpha zero would be fun to watch as it values positions in % on how it would be doing vs itself
This was a great game! Both of those moves you highlighted were spectacular!
One of the advantages of a computer playing chess is that it never gets tired, never gets emotional and never has a bad day.
Even though this was a d4 opening I was still fully ready to hear “and in this position b4, the Evans Gambit, was not played.”
This new merchandise is so nice! Keep that style!
19:17 I think what needs much processing power is train AZ, but running it on any given match should need less than todays engines.
Great idea for the video title. Tal's resurrection...sounds wonderful .
@13:09
Why wouldn't the same idea work if the rook was on a7?
It seems to me it was just a positional advantage, defending the rook with the passed pawn and attacking black's pawn. But regarding the knight moving to g4 nothing seems to be different.
Alpha's move on 13:45 would never be possible if not for the c7 move
White's 26.e5 followed by 27.f5 was known as the "Sweeper, Sealer" back in the day.
that's how a Tal move to brighten up my day..
thanks Agad.
i run it on stockfish 15 and i was amazed! I read everywhere the new stockfish would destroy the old alphazero, and with the advancing of computing and programming i thought it makes sense.
Actually even the most recent stockfish is still blind about that pawn move.
By the way would someone explain me how rook to c2 changed anything? in the previous position with the rook on c7 the knight apparently could still simply jump to f4 but instead the rooks would just take the knights according to Agad.
Why is that?