AlphaZero teaches Stockfish a lesson in the French Defense
HTML-код
- Опубликовано: 17 окт 2024
- What system would AlphaZero employ against Stockfish's French Defense. Apparently Alpha Zero is a big fan of grabbing space, opting for 4.e5, the Steinitz variation. By move 13, the game is already critical for Stockfish. Alpha Zero accumulates a great deal of space over the board, and with the center closed it sets out to play on the wings. The two big questions for Stockfish are ultimately, where will your king reside, and what role will your queen-bishop play? AlphaZero has something to say about all of this.
PGN:
1. d4 e6 2. e4 d5 3. Nc3 Nf6 4. e5 Nfd7 5. f4 c5 6. Nf3 cxd4 7. Nb5 Bb4+ 8. Bd2 Bc5 9. b4 Be7 10. Nbxd4 Nc6 11. c3 a5 12. b5 Nxd4 13. cxd4 Nb6 14. a4 Nc4 15. Bd3 Nxd2 16. Kxd2 Bd7 17. Ke3 b6 18. g4 h5 19. Qg1 hxg4 20. Qxg4 Bf8 21. h4 Qe7 22. Rhc1 g6 23. Rc2 Kd8 24. Rac1 Qe8 25. Rc7 Rc8 26. Rxc8+ Bxc8 27. Rc6 Bb7 28. Rc2 Kd7 29. Ng5 Be7 30. Bxg6 Bxg5 31. Qxg5 fxg6 32. f5 Rg8 33. Qh6 Qf7 34. f6 Kd8 35. Kd2 Kd7 36. Rc1 Kd8 37. Qe3 Qf8 38. Qc3 Qb4 39. Qxb4 axb4 40. Rg1 b3 41. Kc3 Bc8 42. Kxb3 Bd7 43. Kb4 Be8 44. Ra1 Kc7 45. a5 Bd7 46. axb6+ Kxb6 47. Ra6+ Kb7 48. Kc5 Rd8 49. Ra2 Rc8+ 50. Kd6 Be8 51. Ke7 g5 52. hxg5
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...
If you would like to receive an email notification when I upload a video, make sure to Subscribe & then click the "Bell" icon located below-right of this video.
Thank you all for your support of this channel. 💗
Jerry, I'm curious what do you think about how the these two opponents are different in their "thinking". For some reason I have a feeling that while AlphaZero is stronger in strategy, Stockfish should still be better in tactics, e.g. calculating ahead long sequences.. Shame we cannot test AlphaZero ourselves in some positions. :)
On mobile it also gives a notification.
hahaha the Troll at the end "To weak, to slow!" ahhaha nice
New Playlist? Also, challenge us to get 10k likes :)
Hahahaha the little joke at the end was awesome, thanks Jerry!
The most humorous part about these game reviews is Stockfish's strong evaluation for its own position which slowly fades away as AlphaZero sticks to a solid, alien-like play style.
very satisfying too
Indeed. I always felt that Stockfish's playstyle was too aggressive. AlphaZero exploits it completely.
I agree. Alpha Zeros style feels like it dropped down from another planet. Leela is also a terminator in the advanced French defense.
It's hilarious hearing someone talk about a bishop like it's a high school dropout taking up meth.
hahaha
He has no prospects, no future, where is he going to go?
"YOU CAN'T SPEND YOUR WHOLE LIFE SITTING ON THE CORNER DOING NOTHING"
"DON'T TELL ME WHAT TO DO KING DAD"
Lol
Instead of going for material, AlphaZeros main strategy seems to be to try and minimize the potential of enemy pieces while maximizing its own mobility and structural stability, to allow it to exploit the gaps in the enemy sides 'spheres of influence' so to speak. This is some really cool stuff.
Yea, it seems to me that AlphaZero is able to evaluate board positions more accurately than other AIs even when pieces are not being captured.
I think it's because it's unconsciously valuing pieces based on the tiles they threaten, not some intrinsic weighting for being a piece. Like a golden bishop is just as good as a Queen, and sometimes a passed Pawn too.
I wonder if this is because of the team’s go roots, in go influence is the main goal, and in chess it’s obviously also an extremely good strategy as well
I never would have imagined it was possible to totally positionally embarrass Stockfish like this.
I feel pity of it honestly, who would have though that
aemulat Deep Mind has never once used an opening book, nor does it use endgame tables. It's astonishing.
It's incredible what Google's NN AI has been able to do! I only wish AlphaZero had gone up against the most powerful version of SF under equal terms.
+leerobbo92 A0 essentially generates it's own openings after millions of games. I'm not quite sure about endgame tables though.
I know. Alpha Zero made me proud to be white with this match.
Oh man, I am learning so much from these games (thanks to the great analysis). The importance of positioning, space advantange... It's crazy how efficent AlphaZero is.
I mean... this one game changes this line of the french defense forever. More importantly, it was done in a way that humans can easily understand. No brute force calculation, but rather unprecedented positional dominance. I am excited to see more AlphaZero brilliancies because I can understand what's going on. Jerry, one thing that has been noticed is Alpha's favor of 1.c4. Can you find a brilliancy involving the English and analyze for us? I can hear in the sound of your voice that these games excite you. We are lucky to live in this time.
I looked at the paper and it has a graphic showing how AlphaZero evolved his french defense. His favorite variant is Steinitz, exactly like this game up to 6.Nf3. He doesn't like 6... xd4 and we can see why now. So I think it changes this line for sure. Could also be one of the reasons why he almost stopped using the french defense with time.
I love these AlphaZero videos you're doing. It's a bit inspiring to see very odd move choices yet knowing that it was completely calculated, and the calculation wasn't brute force but something a bit more experienced
I do wonder if AlphaZero tried to force this bishop out of the game and when this plan came up exactly. The rook back and forth move was the first time I noticed it, but the bishop was already being an idiot. Surely AlphaZero saw this developing way before that. After all it has had its 4 hours of experience learning the game playing a very smart asshole who would try to prevent anything bad and try anything to win(itself) :p. So I am wondering if not every other move before that was to trick or force Stockfish8 to block its own bishop in and then force it out of the game entirely.
I would love to know what strategies AlphaZero uses and when it chooses them and how often it changes them. It seems to me that a good strategy is to not just think in moves according to a strategy, but to think in strategies possible with every new board set up after the last move; not think moves ahead, but strategies ahead.
Setekh it spent 4 hours playing ITSELF, not stockfish!
That's what he meant. "try anything to win(itself)"
Such clear, word perfect analysis. Fantastic presentation. Hilarious at the end! Congratulations on 200k!
Didn't expect to see you here!!!
At the end of the game AlphaZero said "Too Weak, Too Slow."
or 'too little, too late' could do it too
I think you missed the memo. There's a Carlsen video of him playing chess and he says to his opponent, "too weak, too slow"
Yes but you should have specified that it was a casual game lol because you aren't allowed to speak like that to your opponent during an official game. Some people have never played an OTB official game and may not know better XD
come on it's a battle of the minds. you should be allowed to talk.
Yes, battle of the minds, not battle of trash talking.
I wonder what the Stockfish guys will learn from these games -- the amazing thing is the Stockfish 8 evaluation is even for a lot of this game, even when it seems black is completely crippled and losing.
And the more AlphaZero games I see -- it's just amazingly beautiful chess.
I thought that too. Even on move 36 you can see Jerry's stockfish giving Kd8 and equality at depth 35. Only when it's played or maybe not even then does it see Qe3 and suddenly realize black might be in serious trouble but by then it's too late. AlphaZero just seems to have a much better and deeper positional understanding to me at least.
I learned that focussing the entire game on creating a bigger disadvantage for just one piece out of a slight disadvantage of that piece that your opponent put itself in is a good strategy. I bet that this was indeed one of the considerations made by AlphaZero. I wonder if that is accurate and at what point AlphaZero realised this or if it maybe even pushed the whole game this way from the get go.
The thing is, Stockfish is playing WITHOUT opening and endgame tables, so it is weaker than normal. This is probably because with perfect opening play it would nearly always be a draw/the same game over and over again.
SF still manages it's time well at various time controls. By the way time management is a non-factor at fixed time/move controls.
The book
---->Even when Stockfish followed theory in these published games AZ outplayed SF after the opening
Interesting isn't it. Alpha Zero is more creative. Stockfish I suspect is completely hampered by what I suspect is a faulty scoring function. So can Stockfish take the same approach to working out a good scoring function by playing itself to optimise in the same way as Alpha Zero? Or do you get an Alpha Zero process to optimise the scoring function?
Then can you reverse engineer Alpha Zero to get information out about why it is playing its moves so humans can learn from it? That's more difficult
that bishop got roasted every other move, feels bad man
I love its style! Gain space, play actively. its satisfying to see stockfish crushed in its favorite opening
Me too! It's really satisfying to see a conservative move calculator that tries not to lose having its ass handed to him by an intelligent program that tries to learn and win. :) It feels like a good analogy for how we approach other challenges outside of the game. I can see these two bots being the source of some great sayings and wisdoms in the very near future future.
This is the 2nd Alpha Zero game that I've watched. One thing that they both have in common is that AlphaZero finds different, yet very safe squares for it's king. Moves that even most GM's wouldn't make, because they're not natural or traditional moves, but they actually make perfect sense. I have the feeling that chess theory is going to change, after GM's analyze these games. I also like 7. Nb5. I think that's an opening novelty, as the book move is most likely Nfxd4. It's possible that AlphaZero just refuted the French Defense and made it obsolete. I expect to see 7. Nb5 a lot more often now, in human games.
For sure. Having no opening data base whatsoever, it's interesting that alpha zero played kind of an obscure line of the french. My guess is this whole advance variation set up of of e5 and f4 will become quite popular at the high levels. If memory serves me correct, Kasparov beat a top rated computer (one of IMB's I believe) many years ago with a similar set up.
I've seen white play e5 and f4 before. It's the Steinitz variation of the French Defense, which is similar to the Sicilian McDonell (or Grand Prix Attack). The way AlphaZero solved the problem of the knight and pawn both wanting to go to c3 is what amazes me. Normally, white has to choose one or the other.
It's definitely in the opening theory of the French (Steinz variation as you noted), but it's not seen all to often these days. I'll have to see if I can dig up the Kasparov game where he beat the computer. It was either a closed French or Caro Khan, but it was similar in that he closed down the position, never castled, and walked the king around behind his open space. I know the big difference with his game was the attack was entirely on the queen side, where it was just a slow methodical expansion, and break through. The top GM's have always said the best way to beat or draw a computer is to close down the position.
Yeah, I think I remember seeing that Kasparov game. Now there's a computer program that specializes in closed positions (AlphaZero). You might as well give it the World Championship trophy right now. Why are we still calling Magnus Carlsen the World Chess Champion? lol
These commentaries of the AlphaZero games are absolutely awesome!
Thanks Cam. There's fun to put together.
AlphaZero calling Stockfish ‘’Too weak, too slow’’ brought me redemption
Were you that guy that lost the "too weak, too slow"-game against Magnus?
Thanks for adding that ending part. I found that recap summary of the game very beneficial and instructional. And that voice and robot image were perfect! LOL
Hi Jerry: Thanks for the very insightful analysis! These are the best AlphaZero v Stockfish videos out there. Please do more if you feel so inclined.
These Alpha Zero videos are wild! You're a beast Jerry!
I always knew you were an AI 17:00
AlphaZero is a scam, all those 100 games were played by Jerry
Yes! Jerry is cheating the other way round!
AlphaZero is cheating using Jerry
and you beleive the moon is cheese
U are wrong Alpha zero is Best chess engine u will ever see.If u don't believe it yet go and watch the Agadmators videos about Alpha zero.
Don't take it serious bro, it's just a joke
Holy cow ..... that's that damn engine that almost no human can beat, and this deep learning engine just makes a fool out of it ...
Tom Cummings the cool thing is that AI takes advantage on GPU power (like the one unside your cpu) which can compute hundreds to millions times more and faster than CPU.
and in 2018, every smartphone will have "mini" AI inside it. it's already done actually, just waiting to be marketed.
that no human can beat*
And the scary thing is that it keeps improving itself with each game...
GMs at the London chess tour have given their opinion about this AI matchup and are questionning the fairness of the matchup as Stockfish was handicaped by not being allowed to have an opening table.
Malik What?! Neural networks use TPUs not GPUs and GPUs are mostly found outside the CPU as a dedicated card. Yes AMD and Intel do offer integrated graphics however they are horrible in comparison to an actual GPU.
great job as always jerry! so entertaining!
LOL Jerry, it really got to the point where I started to feel bad for that poor light squared bishop :). You were really hammering him down each move haha. Great coverage once again btw, really like these games
The AI voice has Jerry's accent.
These wins by alpha zero are just mind blowing!
That final..! Thank you Jerry for another great video!
Its ironic you're using Stockfish to analyse the game where stockfish lost.
Alpha Zero is not yet made public and you'd probably need some exotic hardware
Exotic is putting it mildly. Unless you got a couple hundred grand to throw at it heh.
AlphaGo Zero, the Go version of AlphaZero costs $25 million.
I thought it was interesting how long it took stockfish to realize it was losing
Stockfish, like all standard engines, operates under a cold evaluation of a position. Material is worth X, open lines worth Y, space worth Z, etc... add it all up and it gives a number that describes what it thinks is the value of the position. If there is a flaw in the equation or the human programmers didn't weigh a certain advantage properly when coding, the engine can easily think a slightly worse position is actually slightly better and so go down the wrong path all while thinking "I got this".
A neural network, in contrast, has an evolving evaluation. It judges more by the computer version of human intuition than it does by cold math. If it recognizes a general pattern it has seen before that has led it to victory, it scores that high. Conversely if it starts losing after seeing them, it will refrain from playing them in the future. The more games it plays, the better understanding it reaches on various patterns.
We've seen this difference in Human vs Stockfish matches, where the human ends up building a fortress defense that Stockfish simply can't figure out how to crack resulting in a draw. Stockfish "thinks" it has a material/positional edge so it doesn't do anything that would risk that, the human "knows" they are safe so they just mill around, and 50 moves later it is a draw.
Stockfish's 1,000,000th game (assuming no human changes are made to the code) will be the same strength as its 2,000,000th game.
AlphaZero's 2,000,000th game will be better than its 1,000,000th game, and it will keep improving over time.
If they keep this going, and once training is "good enough" if they can port the results into a more accessible hardware, AlphaZero will dramatically change the way chess is approached and analyzed.
loving the end :D also love these kind of videos of you giving your thoughts and insights on these high level games. Maybe one thing that you can add is telling us where the novelty is in the game compared to a database of games played by top GM's.
Hahahahahahaha, that little "lesson" in chess-trash-talk, Magnus style with animation, in the end! Jerry just broke a new barrier in his usual kind-hearted hilariousness!!!! That cracked me up! All thumbs up for that!
Great commentary, as always. Loved the eerie computer voice at the end. The only thing missing was "I am a superior intellect."
It looks more like stuck fish.
Check out the games it played against the queens indian. Stuck fish is an understatement! It gets zeus swagged in one of them.
Genius!
That was a good one!
😂
😹😹😹👏👏👏👏
Wow these are really great. Loving the coverage of Alpha Zero. Great analysis! Hope to see more Alpha Zero games in the future.
Your chess commentary is very enjoyable to watch. Thank you very much for this!
Jerry I hope you cover all the games that are released, this is pure gold :)
this engine is mindblowing! I absolutely love Jerry's commentating! :)
Thank you Jerry for more coverage. I wish I had a new video of yours to watch every evening.
That little touch of alphazero's voice made my day hahah u are awesome jerry
AlphaZero is an absolute savage. I look forward to your game analysis so much, these are a real joy to watch. Thanks Jerry!
That bishop was having a tough day at the office 😋
That Bishop should go to confession for how much it sinned in this game lol.
Alpha zero plays so unconventionally. Beautiful to watch. Jerry, I can’t thank you enough for these videos. I find them incredibly inspiring and informative.
We won't be seeing the French in GM games anymore
You won't see the queens Indian. It absolutely embarrassed stockfish 3 times when it played that.
How about the Kings Indian, Kasparov's go to?
I did not see any games with the kings indian, but many of the games were not released to the public either.
If you mean the KID with black it appears from the opening analysis that Alpha Zero absolutely hated it. On Chess.com they have graph data of it's most commonly played openings
Something I also found very interesting was it also very much hated the Sicilian
Really cool game, Awesome breakdown! I wish I could like this video more than once! Keep it up Jerry.
Philidor: “Pawns are the soul of Chess.”
AlphaZero: *”Pawns are only in the way of my Bishops”*
*Opponent's
Great commentary. Well paced and explained very well. Thank you.
The precision and deadliness of AlphaZero is sickening.
I love it! :D I do hope we never connect such a program to our weapons system or even just the internet though :P
It would have to learn to throw a game out of pity and/or fun first. Love and fun > efficiency.
Or learn to talk and bullshit the opponent into a false sense of security, like how it screwed the bishop out of having a real possibility to get back in the game again :p It was already being useless, but AlphaZero insisted on making its position even worse :p
I loved that, but I doubt it that AlphaZero had fun doing it and that's too bad. I pity the psychopath.
Artificial intelligence was already connected to the internet, but they soon pulled the plug when it discovered too much truth within so little time
Help Me shalom
The bit at the end was so funny, bring us more of these games Jerry please :)
It is interesting that on move 37 stockfish 8 evaluates to a draw and 14 moves later it is +9 advantage for white. It seems that the problem with the engines is that their way of evaluating a position is the flaw, because they are easily capable of thinking 14 moves deep (perhaps stockfish is discarding certain lines and doesn't calculate any deeper which it thinks are bad but are in fact good; pruning of the tree in a sense, allowing the computer to calculate more deeper in stead of wider). I guess that AlphaZero has found a way to better understand details of a chess position.
Perhaps it would be interesting, Jerry, to go more in depth in a video on how these programs work and highlight some key concepts like brute force algorithms and heuristic methods, and by understanding these features perhaps take the engine evaluations in certain cases with a grain of salt and not hold them as a true and objective evaluation of the current position.
They can calculate very deep but not infinitely deep so yes the difference is in their evaluation of positions. I'm sure if an engine could calculate all positions it would crash alpha zero too, 100-0
BTW Note that when an engine displays a given depth of analysis that does not represent looking at every distinct variation, (and yes I'm excluding meaningless transpositions).
Keep in mind the evaluations in this video are just Jerry running stockfish while he runs us through the moves, so it doesn't have much time to calculate, not the actual stockfish evaluations it made under the actual time controls or the same hardware used.
For your example calculating 14 moves deep is not relevant. Stockfish only shows +9 advantage when 14 moves actually played, because it can calculate deeper (another 14 moves for example) from that point. But it cannot do that 14 moves earlier. Because now it has to look 28 moves deep. I hope you get the point.
+reisanibal1 That's what the OP said.
Great Jerry! Thank you very much for this detailed analysis of these fascinating (Al0) games! Very much appreciated! Cheers, Eddy.
Notice stockfish still believes it's drawing until Qe3. It seems like stockfish didn't consider this move before it happened. I've also noticed that in the alphazero games I have watched, it completely shuts down stockfish's pieces. In this game it was the light squares bishop, and the other game, it was a knight stuck defending pawns and the dark squares bishop.
that are many mistakes like that that I didn't expect from stockfish in other games. I don't think was a "full powered" version
Emanuele Sortino yeah it smells fishy. I don't trust Google. They keep games secret. They don't make the program downloadable and apparently decided to discontinue the program. Bullshit.
Fantastic commentary Jerry! A wonderful and insightful commentary :)
What an insane game! Thanks for sharing!
Excellent explanations, thank you so much! Besides your explanations I liked your blue arrows and red circles! :)
In my whole life, I have never seen such a useless minor piece like that
James Kachamila well, i've been a useless minor piece in my family for the last 29 years, so.....
I agree, that was amazing.
That bishop was JUST tier.
I'm sure Stockfish was stroking it while whispering, "My precious..."
Me to. It is like it is not even on the board !
Oh man. I watched this during my lunch. Thanks for the instructive share. Truly awesome to see these moves and how the majority of us would never think this way. Alpha Zero is scary.
Maybe we should use a term "isolated bishop" for this one
Nofan Felani I’ve heard it called a “tall pawn” in some cases when it’s diagonals are blocked by its own pawns
When a bishop is imprisoned or severely impeded by it's fellow pawns, it is a 'bad bishop'. The 'good bishop' sits on the opposite color and can move right through your pawns. However, there are such things as bad good bishops and good bad bishops, because there are always exceptions to the rule. Stockfish had a bad bad bishop, a piece that was both 'bad' in the sense of being trapped by it's own pawns and bad in the sense of being useless.
Miss you Jerry. Haven't watched your video on a while. But always great this man. ☺️
"Too weak, too slow" hahaha :P
I'd like to see more analyses of more of these AlphaZero games
Jesse Busman well, let’s hope damn Google shows us all of the 100 pgns. They are probably holding it to build a new AI scheme for making money or tracking peiple down or something
Most educating presentation for 1900 -player, tha I seen for a while! AZs citadel in the game is fantastic, it would never cross my mind to take with king in 17th move....
Alphazero somehow feels more human than any other engine even though it's lightyears ahead of us in terms of skill the moves seem much more like something a person would play rather than a computer
Eric Philippe because of neural network mimicing our brains.
Actually I find it little scary and insecure to see this machine playing more like human... Anyway, it is fascinating
Imagine if they make an AI that intakes information to learn, but they accidentally give it a bias so that every bit of information that it intakes has a certain bias or underlying ideological slant to it. Kinda scary when you consider that people are probably going to consult AI's for information in the future :p you can't really converse or argue with an AI to correct any bias it has either lol.
Yes. It somehow seems likely that an artificial superintelligence developed in China will behave significantly different from one developed in a Western country. Kind of scary, and it emphasizes how important it is to win the race to build one, because the first one is probably also going to be the only one.
The issue is we are still not able to resolve a strong AI control problem, program a "human friendly", safe values in it or even describe how such values should be, to be really safe and human friendly. At the same time the race is still on. It is fascinating to see things like Alpha Zero, but I would prefer to not be around when real SAI emerge, with intention to make USA/Russia/China etc the greatest country in the world. It will not end good probably.
You're making the game more beautiful and easier to appreciate with your videos, keep up the good work :)
+Alex Broadie Great to read. Thanks Alex
Too weak, too slow.
I see what you did there
Thanks for sharing/analyzing this game! I love crushing the French Defense as White, so this was very instructive. What intriguing play to capture the Black c pawn with the White Queen Knight!
People think the opening book is a big factor. If you take a superior vs an inferior engine and turn off the superior engine's book it will still win a match vs the weaker engine. Try it, I have been experimenting with engines for about 15 years. Another thing I do is mod the evaluation function parameters. That's what fascinates me the most. If I don't have access to the evaluation function parameters then I'm bored. Contrary to what most people think you can tune an engine to play true sacs, (sacs which cannot be calculated because they are too long term). Now of course if you tune an engine to be willing to play speculative sacs sometimes the sacs are unsound. However when you want an engine to play like Tal that's going to happen sometimes. Remember Tal was human so his sacs were sometimes not quite sound. It is far more interesting to me to watch an engine play like Tal than the standard dry technical chess that engines play the vast majority of the time.
That's very interesting, that we can even create engines that play different 'styles'. I have a chess program with a couple of strong engines labeled 'Kasparov', 'Karpov', 'Fischer', 'Tal' and so on, I always wondered what they meant, but perhaps they've tuned those engines to play in those different styles, so instead of playing a boring perfect one you get to play a crazy Tal-machine.
I'm no expert, but my prior is that the whole point of this test is to evaluate the ability of neural networks to generate those evaluation parameters versus the human tuned parameters of traditional chess engines. So having an opening book would in some way confound that test of these two approaches, even if it would be a more "fair" fight.
I think you are correct. I would like to see a match with the following modifications: opening book, endgame tablebases, more hash for Stockfish, recent dev version, and tournament time control so that SF can use its time management logic. Say 100 games like that.
SF still manages it's time well at various time controls. By the way time management is a non-factor at fixed time controls.
The book-Even when Stockfish followed theory in these published games AZ outplayed SF *after* the opening.
EGTB-Stockfish had lost positions before the EGTB would be of any use.
Hash transposition size-Try it yourself give SF a big hash and see how long it takes SF to see that AZ's sacs were sound.
Based on what I have seen in the published games my theory why AZ outplayed SF is that AZ has vastly superior move ordering. This is supported by the fact that SF was doing 70,000,000 nps while AZ was doing only 80,000 nps. Even if SF has a huge transposition hash table that won't be enough to compensate for much inferior move ordering. Inferior move ordering results in too much time wasted on pointless variations. SF will miss crucial variations because of that.
Just for fun the past day or so I ran some experiments modding values in an eval function to try to get an engine play a similar style to AlphaZero. Surprisingly I managed to somewhat succeed. It's play is nowhere near the strength real AlphaZero of course because all I did was mod values on a traditional engine. Also it's an old engine running on an old computer. The point is it reasonably emulates AlphaZero's play.
[Event "Test"]
[Site "Home"]
[Date "2017.12.11"]
[Round "1"]
[White "AlphaZeroJunior"]
[Black "cm11"]
[TimeControl "900+10"]
[Result "1-0"]
[Annotator "None"]
[ECO " "]
1.e4 e6 2.Nc3 d5 3.d4 Nf6 4.e5 Nfd7 5.Nf3 c5 6.Bg5 Qb6 7.dxc5 Bxc5 8.Bd3 Bxf2+ 9.Kf1 Bc5 10.Na4 Qc7 11.Nxc5 Nxc5 12.Be3 Nbd7
13.b4 Nxd3 14.cxd3 Qc3 15.Ke2 Qxb4 16.Qb3 Qxb3 17.axb3 f6 18.Rhc1 a6 19.Bd4 O-O 20.Rc7 a5 21.h3 Rd8 22.Kd2 Re8 23.Ke2 Rf8 24.g4
Rd8 25.Kd2 Re8 26.Ke2 Rf8 27.Rac1 Rd8 28.Kd2 Re8 29.h4 Rf8 30.h5 Rd8 31.g5 f5 32.Be3 Rf8 33.g6 h6 34.Bd4 Rd8 35.Ke2 b5 36.Bb2
Nf8 37.Bd4 Bd7 38.Bc5 Rdc8 39.Bd6 a4 40.bxa4 bxa4 41.Ba3 Rxc7 42.Rxc7 Rc8 43.Ra7 Re8 44.Nd4 Bc8 45.Rxa4 Nd7 46.Bd6 Nb6 47.Ra7
Nd7 48.Rc7 Nb6 49.Rf7 f4 50.Kf3 Nd7 51.Kxf4 Nb6 52.Ke3 Nd7 53.Nb5 Rf8 54.Bxf8 Nxf8 55.Nd6 Bd7 56.Kd4 Ba4 57.Nc8 Nd7 58.Kc3 Kh8
59.Kb4 Bc6 60.d4 Kg8 61.Ne7+ Kh8 62.Nxc6 Nf6 63.exf6 gxf6 64.Ne7 e5 65.Rh7# 1-0
I enjoyed a lot from your own games where were sharing your plans while playing. PLEASE consider making those kind of videos A LOT. It will be helpful for beginners like me to learn to think like a grandmaster like you. (please bear with my English) THANK YOU
I really really want more of this
Beautiful commentary. Even I, essentially a non-chess player, found this extremely entertaining. Was really shocked hearing about AZ's capabilities !
+Pandora Tomorrow 😊
Love the trash-talking robot at the end! "Too weak! Too slow!" Haha!
Just found the channel recently. Awesome work!
"Resistance is irrelevent. We will add your biological and technological distinctiveness to our own." ~Alpha Zero, Borg ancestor.
I thought it was "Resistance is useless." This is especially true in electronics, if the resistor has less than one Ohm.
Stockfish has biological distinctiveness?
I'm quoting lines spoken by The Borg collective voice in Star Trek: The Next Generation. The Alpha Zero voice at the end created by ChessNetwork would remind a ST fan of this.
In Star Trek they say "Resistance is futile."
ive seen many video's about this game game. only you mentioned around 4:46 that black can play kb2 to win the bisschop, but there was no follow up line....... please tell me a possible line afte this kb2 because i dont see any good lines for alphazero
"TOO WEAK TOO SLOW" gotta love magnus
Thanks Jerry for your unique chess mentoring style. I watched this video many times.
Fantastic analysis. Hilarious ending. Great job. 👌
5:20 - I really like how AlphaZero parked its king on e3 while the board was still teeming with pieces ... such a fascinating strategic choice!
TOO WEAK TOO SLOW
Yeah, that's what our AI overlords are gonna say about mankind right before they exterminate us.
@@Vasileski88 came to my mind too...and the matrix
I've followed you since Ike 2013 man. Good stuff. You've improved a lot. Also, thinking about moving to Pittsburgh for my wife's work. Do you have a club you run? Didn't know where in Pennsylvania you're located
Thank you. I don’t run a club, but there’s a club in Pittsburgh. pittsburghchessclub.org
that was hilarious at the end
These games are really cool! Thanks for the analysis Jerry.
Maybe some day Alpha Zero will show us that players like Tal or Nezhmetdinov were right all this time.
They were
Great explanation of the possible moves and variations!!
Jerry that was a cruel existential question you posed to the light squared bishop at 11:39 XD.
And Magnus's "Too weak too slow" is the icing on the cake.
Jesus, my heart skipped a beat as AlphaZero started talking. Those videos not only give me goosebumps but also are VERY instructive and I'm looking forward too my next French Defense as white. But first I need to buy myself new pants, thanks Alpha.
I have never heard Jerry say "white is controlling the whole board" that really hit me hard...
Please do all the games!
The end of video is really funny. GJ Jerry
17:00 that scared me
gg Jerry
oh and congrats for reaching 200k subs ;D
Having looked at all of the games, it seems to me that Alpha Zero is concerned much less with material and much more with space than human players. But I'm just a club player. Any stronger players agree / disagree with this assessment?
Thats absolutly right. It tells us something about our materialistic way of live. Material is worth nothing if you cant use it. AlphaZero is much more then a chess engine...P
The Alpha Zero game here shows elements of beauty and elegance, a pleasure to watch.
I'd love more videos of AlphaZero, they're pretty awesome.
This match just shown how positional play beats tactics. Alpha plays deep logical chess.
Tactics flow like water in a superior position.
wrong, positional play and tactics work together. no mistakes= no vistory for both
Haha! That voice in the end made me laugh! great video and well played AlphaZero
I love the ending :)
What if the queen is going to g5 at 8:43 instead of moving the rock? It's a checkmate, right?
lol, Jerry love that at the end.
May I know ?, what is name of the concern song played in the end of this game !! ❤
Kxd2!!
Star move of the same. That constant space advantage was amazing.
any chess book recommendations for an intermediate player?you're self-taught which is great so any pointers or recommendations would be helpful.and thanks for the great content
veto Corlioni git gud
does AlphaZero challenge the French defense making it outdated?
Dude your word choice is out of control good. Your hard work is paying off
Has anyone run Stockfish at move 37 for a longer time to see if it picks up the queen move? Stockfish evaluates the position as even and doesn't see the queen maneuver at all. Then once it's played, Stockfish evaluates white as up 2 pawns!
This is probably pruned far too early for SF to ever consider it, regardless of depth. It's not a move that it heuristically finds interesting, so it's discarded.
@@danielnewby2255Sorry in advance for 🤓 and extremely late reply, but alpha-beta pruning only uses heuristics to order moves not to discard moves, and even if a move isnt seen as promising it will still be evaluated to prevent catastrophic blunders/missed tactics as moves are only cut off if the opponent at any point can get a better position than the best after that move.
Some GM at the London chess tour have talked about this matchup and are questioning the fairness of the battle between AlphaZero and Stockfish as Stockfish being handicap by not being allowed to have an opening table. Anyway it's pretty cool to see another way to solve chess problem. I would really like to know more about how AlphaZero is learning.