Alphazero's shortest loss to Stockfish | Shocking French Defence Classical secrets revealed
ฝัง
- เผยแพร่เมื่อ 9 ก.พ. 2025
- ♚COURSES kingscrusher.t...
📚 kingscrusher.t...
📚 kingscrusher.t... || kingscrusher.t...
#KCComputerChess
#KCAlphaZero
♚ Play turn style chess at bit.ly/chessworld
♚ Play Chess vs. Kingscrusher and others: www.chessworld...
♚ Subscribe to best TH-cam Chess Video Channel : bit.ly/kingscru...
FIDE CM Kingscrusher goes over amazing games of Chess every day, with a focus recently on games of Neural Networks which are opening up new concepts for how chess could be played more effectively.
👕 Kingscrusher Store: teespring.com/...
♚Donating via Paypal: goo.gl/7HJcDq
♚Channel exclusive perks: / @kingscrusher
Game quality tags: amazing, awesome, astonishing, brilliant, classic, crushing, dynamic, elegant, exceptional, excellent, exciting, fabulous, famous, fantastic, finest, flashy, greatest, important, impressive, incredible, instructive, interesting, magnificent, marvellous.
Info about Leela Zero:
en.wikipedia.o...
...
Leela Chess Zero (lc0) is a free, open-source, and neural network-based chess engine and distributed computing project.
Leela Zero's algorithm is based on DeepMind's 2017 paper about AlphaGo Zero.[3][6] Unlike the original Leela, which has a lot of human knowledge and heuristics programmed into it, Leela Zero only knows the basic rules and nothing more.[7]
Leela Zero is trained by a distributed effort, which is coordinated at the Leela Zero website. Members of the community provide computing resources by running the client, which generates self-play games and submits them to the server. The self-play games are used to train newer networks. Generally, over 500 clients have connected to the server to contribute resources.[7] The community has provided high quality code contributions as well.[7]
Leela Zero finished third at the BerryGenomics Cup World AI Go Tournament in Fuzhou, Fujian, China on 28 April 2018.[8]
Info about Alphazero:
en.wikipedia.o...
AlphaZero is a computer program developed by the Alphabet-owned AI research company DeepMind, which uses an approach similar to AlphaGo Zero's to master not just Go, but also chess and shogi. On December 5, 2017 the DeepMind team released a preprint introducing AlphaZero, which, within 24 hours, achieved a superhuman level of play in these three games by defeating world-champion programs, Stockfish, elmo, and the 3-day version of AlphaGo Zero, in each case making use of custom tensor processing units (TPUs) that the Google programs were optimized to make use of.[1] AlphaZero was trained solely via "self-play" using 5,000 first-generation TPUs to generate the games and 64 second-generation TPUs to train the neural networks, all in parallel, with no access to opening books or endgame tables. After just four hours of training, DeepMind estimated AlphaZero was playing at a higher Elo rating than Stockfish; after 9 hours of training, the algorithm decisively defeated Stockfish 8 in a time-controlled 100-game tournament (28 wins, 0 losses, and 72 draws).[1][2][3] The trained algorithm played on a single machine with four TPUs.
...
Relation to AlphaGo Zero
Further information: AlphaGo Zero
AlphaZero (AZ) is a more generalized variant of the AlphaGo Zero (AGZ) algorithm, and is able to play shogi and chess as well as Go. Differences between AZ and AGZ include:[1]
AZ has hard-coded rules for setting search hyperparameters.
The neural network is now updated continually.
Go (unlike Chess) is symmetric under certain reflections and rotations; AlphaGo Zero was programmed to take advantage of these symmetries. AlphaZero is not.
Chess can end in a draw unlike Go; therefore AlphaZero can take into account the possibility of a drawn game.
AlphaZero vs. Stockfish and elmo
...
►If you like my chess videos please consider become a full member here at my Chessworld Site. www.chessworld....
#KCChess game tactics
♞ Challenge KC and others for turn style chess at bit.ly/chessworld
♚TEAM kingscrusher.tv...
♚TWITTER kingscrusher.tv...
♚DISCORD kingscrusher.t...
♚TWITCH / kingscrusher
♚REDDIT kingscrusher.tv...
♚QUORA kingscrusher.tv...
♚DONATE paypal.me/king...
♚COURSES kingscrusher.t...
♚COURSES kingscrusher.t...
Replayable game with indented variations: www.chessworld.net/chessclubs/ltpgnviewer32/ltpgnboard.asp?GameID=5024043
We know why Alpha discarded the French in 1 hour I think
Fun comment - Cheers, K
... which means that A-0 hasn't practiced this defense.
It means the frech defense sucks and is bad?
@@AAYLV Alpha Zero doesn't do French Defense naturally. It was forced to do French Defense in this game. This means that French Defense is an inferior defense with very slim possibility of success in the long run.
@@KevinJDildonik , I agree with your first comment, but not the rest. AZ always plays the best move possible in any position. It rejected the French simply because the French is not a good opening like every one thought.
Can you please cover the reverse game? I'd like to see how SF tried to defend this.
+
alpha sacs a piece in the attack and the game draws after 173 moves. not as entertaining
The French Defense was among the ten openings analyzed by Luis Ramirez de Lucena in 1497. It was played in the last game of the Morphy-Anderssen match in Paris in 1858 and also in the 1960 Fischer-Tal Olympiad game in Leipzig which ended in a draw. This Stockfish-AlphaZero game is of course out-of -this-world !!!
It seems both Stockfish and Alphazero agree that the Alekhine-Chatard Attack is sound.
The other hilarious thing about that Bc4 move is that Stockfish then sets up the exact thing that you should try to normally avoid which is a fork by a pawn on two minor pieces. I mean that's exactly what I spend time trying to avoid, and here Stockfish actually deliberately sets up the two minor pieces so that the pawn is forking them. Hilarious!
French DeShit, twice that opening but still kudos to the Fish for managing to exploit it so well.
The French defence is still played at the top levels of Over the board Chess but does sometimes lead to overly cramped positions. Cheers, K
Could you please be clear on which moves were book moves and which were engine moves? Add it to the description perhaps. It's very confusing to me, not knowing when AlphaZero starts playing.
That is why AlphaZero refuted this opening! AlphaZero wouldn't play the French Defense because it does not see it as a good opening.
Cheers, K
But it just DID play the French as seen in the video....u okay broo???
@@Cris-qg5fk That's because it's book moves, which are moves not picked by the engines themselves. It's like you are forced to play an opening you hate.
This is the difference between a good engine, and the best engine. Stockfish can draw a horrible opening and win with tiny edges in the opening, where as other engines would lose and draw for example.
Which means Alphazero is very weak outside of the game tree it plays itself, suddenly blind to tactics easy for the superstrong traditional engines.
The trend that I am seeing is that Stockfish only wins when Alpha has garbage openings like the French and Chigorin forced upon it.
but alpha only drew this opening vs sf as white. That being said, these softwares are outdated now anyway, both sf and probably alpha zero as well are considerably stronger now
Then don’t.
Not the case. Stockfish with black drew the game. LPHA ZERO is better at positional play, stockfish is stroner in tactical and dynamical positions. also alpha zero didn't play under official tcec rules which gave it some advantage.
If you check tcec archive you can find a very similar game with Ng5 won by SF7 I believe vs Komodo in tcec final the game is identical until it's completely lost.
I sometimes find myself getting into these closed complex positions in over the board play (online and real) and the problem is always the clock. There's just not enough time to calculate everything, even though you intuitively can see from the position that you should have an advantage.
Bishop kaboom c4!! If someone would play that against me I would be thankful for my losing part in an immortal game! Seriously.. Bishop c4.. LMAO! Insane!!
Cheers, K
Yup! And I'd never find that move if I had a year to think about it.
I mean if you force Alpha play moves that it would never play because in its training Alpha learned itself that they were bad moves, then the loss is expected.
That's pretty cool: attacking d5 with the Bishop to c4. Never seen anything like it, really, that's supposed to be the 'impregnable' black central base.
That e5 pawn is not a fawn pawn but its the same concept...not as strong because its on the 5 th rank but still strong since it's in the middle of the board..
this is an insane move. if mikhail tal sees this he will be overjoyed
At 11:51, wouldn't nh7 by black neutralize white's attack? Not only would it prevent the king/rook fork but it would also protect the g5 f6 squares. Yes the rook could continue the attack with h5 capturing the pawn and threatening the knight, but you could push g5 to keep the queen from penetrating? That knight on e4 is still a deadly threat no question, so this may not work once the rook takes the knight on h7.
I definitely think that neural net chess systems will exceed traditional engines at some point, but this may take a few more years yet. Ultimately though, I foresee hybrid systems using both strategies. If both strategies give the same next move, then just play it. If they disagree, then how do you program choosing between the options. But how often would they disagree? Perhaps as infrequently as one move in 20. Interesting times for chess playing specifically and for problem solving more generally.
I think it depends on the sharpness of the position. The neureal network does a monte carlo search which is probabilistic in nature. If there is a single specific line which is a loss, it could miss it.
Did Stockfish fair any better with the black pieces in this opening?
I just noticed he's not really at the beach.
frozencloud17 Ha.
In the earlier days the French Defense was not even popular. Lots of brilliancies won by White. It's a counter attacking opening actually which Alpha zero failed to execute in this game.
Just when I was starting to feel sorry for poor o'l Stockfish... As an underdog it's so much more attractive.
Honest question, why would they use an opening book for the first moves? Is it to get different dynamics?. Would of thought it would be better with no opening book🤔.
It helps to test the engines in different conditions, trying to gauge their overall strengths
It's just a stupid thing to do. You know Alpha would never use that opening so it's a pointless game, really. Still though, thanks KC!
Well, for chess players who play this opening, seeing how AlphaZero deals with the position, may help improve their own game
Plot twist: Alpha ALLOWED Stockfish to win out of sympathy.
did all of alpha zero's losses come when it was forced to play certain moves?
123 456 I know I’m late, but yes
This appears to have been a new flawless game from Stockfish. Always come back to that question regarding the 3400 machines, would you plan on playing the same sort of game if you were playing Mikhail Tal or Anatoly Karpov?
So I guess an obvious question is did Alphazero get white in this position and how did that game go?
I am not quite sure - maybe someone can find it here: deepmind.com/research/alphago/alphazero-resources/ - I couldn't find it myself. Maybe they didn't play it the other way round. Cheers, K
deepmind.com/documents/256/alphazero_stockfish_tcec_positions_all.zip
[Event "Computer Match"]
[Site "London, UK"]
[Date "2018.01.18"]
[Round "11297977657538292528"]
[White "AlphaZero"]
[Black "Stockfish 8"]
[Result "1/2-1/2"]
1. e4 { book } e6 { book } 2. d4 { book } d5 { book } 3. Nc3 { book }
Nf6 { book } 4. Bg5 { book } Be7 { book } 5. e5 { book } Nfd7 { book }
6. h4 { book } Bxg5 { book } 7. hxg5 { book } Qxg5 { book } 8. Nh3 { book }
Qe7 { book } 9. Qg4 g6 10. Ng5
h6 11. Bd3 Nc6 12. Nxf7
h5 13. Qg3 Kxf7 14. Bxg6+
Kg8 15. Bxh5+ Qg7 16. Qf4
Ndxe5 17. dxe5 Qxe5+
18. Qxe5 Nxe5 19. f4 Ng4
20. Kd2 Nh6 21. Rae1 Bd7
22. g4 Rf8 23. Ne2 Nf7
24. Nd4 Nd6 25. Nxe6 Bxe6
26. Rxe6 Rxf4 27. Rhe1 Ne4+
28. Ke3 Rf8 29. Kd4 c6
30. Re2 Rf4 31. Ke5 Rf8
32. c4 Ng3 33. Rg2 Nf1
34. cxd5 Ne3 35. Rg6+ Kh7
36. Rh2 Nxd5 37. Rg5 Kh6
38. Rf5 Rxf5+ 39. gxf5 Kg7
40. Rg2+ Kf8 41. Bf3 Nb6
42. f6 Rh3 43. Bg4 Re3+
44. Kd6 Nd5 45. Rf2 Rd3
46. Bf5 Rd1 47. Bc2 Rd4
48. Ke5 Rh4 49. Ke6 a5
50. a3 Rh3 51. Bg6 Rg3
52. Bb1 Rb3 53. Ba2 Rd3
54. Bc4 Rg3 55. Kd6 Kf7
56. a4 Kf8 57. Ke6 Re3+
58. Kd6 Rg3 59. b3 Ne3
60. Be2 Nd5 61. Bc4 Ne3
62. Rh2 Nxc4+ 63. bxc4 Kf7
64. Rh7+ Kxf6 65. Rxb7 Ra3
66. Rb2 Kf7 67. Kd7 Kf6
68. Kd6 Kf7 69. Kd7 Kf6
70. c5 Rxa4 71. Kxc6 Ra3
72. Kc7 a4 73. c6 Rb3
74. Re2 a3 75. Kd8 Rc3
76. Kd7 Rd3+ 77. Kc8 Rc3
78. c7 Rb3 79. Kd7 Rd3+
80. Kc6 Rc3+ 81. Kb6 Rb3+
82. Kc5 Rc3+ 83. Kb6 Rb3+
84. Ka7 Rc3 85. Kb7 Rb3+
86. Kc6 Rc3+ 87. Kd7 Rd3+
88. Kc8 Rc3 89. Ra2 Ke7
90. Rd2 Rb3 91. Ra2 Ke6
92. Kd8 Rd3+ 93. Kc8 Kd6
94. Rc2 Rg3 95. Rd2+ Ke6
96. Re2+ Kd5 97. Kb7 Rb3+
98. Kc8 Rb2 99. Kd7 Rxe2
100. c8=Q a2 101. Qc1 Rc2
102. Qd1+ Ke5 103. Qg4 Rd2+
104. Ke8 Rc2 105. Qa4 Kd6
106. Kf7 Rf2+ 107. Kg6 Ke5
108. Kh5 Rb2 109. Kg5 Rc2
110. Kg6 Rf2 111. Kh5 Rb2
112. Qa3 Rh2+ 113. Kg4 Kd4
114. Kf5 Kc4 115. Ke5 Rd2
116. Ke4 Re2+ 117. Kf4 Rf2+
118. Kg5 Rc2 119. Kf5 Rd2
120. Kf6 Rc2 121. Ke6 Rd2
122. Ke5 Rc2 123. Kd6 Kb5
124. Kd7 Rd2+ 125. Kc8 Rc2+
126. Kb7 Rh2 127. Ka7 Re2
128. Kb8 Rd2 129. Qc3 Ka4
130. Kb7 Rh2 131. Kc6 Rh6+
132. Kc5 Rh5+ 133. Kc6 Rh6+
134. Kc7 Rh7+ 135. Kb6 Rh4
136. Kc5 Rh5+ 137. Kd6 Rh3
138. Qb2 Rh2 139. Qxh2 a1=Q
140. Qc2+ Ka3 141. Qc5+ Kb3
142. Qd5+ Kb2 143. Qd4+ Kb1
144. Qd3+ Kc1 145. Qc4+ Kd2
146. Qf4+ Ke2 147. Qh2+ Kd3
148. Qh3+ Kc2 149. Qf5+ Kc1
150. Qf4+ Kb1 151. Qe4+ Ka2
152. Qc4+ Kb1 153. Qf1+ Ka2
154. Qf7+ Kb1 155. Qf5+ Kc1
156. Qf1+ Kb2 157. Qg2+ Kb1
158. Qh1+ Ka2 159. Qd5+ Kb1
160. Qe4+ Ka2 161. Qe6+ Kb1
162. Qg6+ Ka2 163. Qf7+ Kb1
164. Qh7+ Kc1 165. Qh1+ Kb2
166. Qb7+ Kc1 167. Qh1+ Kb2
168. Qg2+ Kb1 169. Qg1+ Ka2
170. Qa7+ Kb1 171. Qg1+ Ka2
172. Qa7+ Kb1 173. Qg1+ 1/2-1/2
When Stockfish played the french defence A0 won easily. Not nearly as quick though.
@@ns9176 Nope.
This just looks like checking the bad openings. 6. ..., Bxg5?. The ? Is already mentioned in Euwe 1965, and it was probably mentioned far earlier. Unless in the meantime some rehabilitation happened, which I don't believe, the only merit to use these openings in computer chess is to see whether the old masters and theoretics were right, and whether there is an even more convincing refutation. AlphaGo, Black, losing this game is not the point, after Bxg5? the position is already lost for sure. The point is, Stockfish, White, winning it, letting it show that it can win a won position, in a convincing way. If AlphaGo is less tactical oriented, then Alphago having White and Stockfish having Black would be more interesting, where AlphaGo must show it can push through. But, here in the game shown, it is not AlphaGo that lost, it isn't surprising at all. In the meantime it is quite enjoyable to see how relentless Bxg5? is refuted.
Probably instead of Bc4, Nce4 works too, if you are going to sac to get to the f6 square this does the same thing, doesn't it? Although maybe better to try to get rid of the light square bishop since knights are more powerful there. I would've played a knight sac to get the job done, since both are winning me thinks, but I like how creative the bishop move is now lol
Dumb beginner question: If after 19...c6 moving either knight to e4 works, why doesn't it work for white on move 19? Relatedly, after the second knight goes to e4, what's wrong with Nh7 for black rather than Kd7?
Alpha still has never lost when playing without any intro rules
But stockfish held draw when played in reverse color. It shows that in extreme tactical and dynamic positions stockfish can see something that AZ can't.
Plus today's stockfish version is stronger by 115 points then stockfish 8!! Means it will probably win against that version of alpha zero today. (unless AZ will make more progress.)
@@holahola-gp6vd no
I had to watch this again to see where A0 went wrong. I think it is early on at bxg5 but if A0 was forced to play that book move, I wouldn’t consider this to be fair. Bxg5 might be a losing move. If Stock fish was playing black, would it bxg5?
the reverse game was played and it was a draw, it's somewhere in the comments above
2:30 turn on cc lol
Do you think Ng5 is almost just refuting this whole line? Nowhere in the video do you really offer any alternative for black and certainly after 13 moves white seems to be winning by force.
Cool comment, K
NN's naturally can't compete with traditional chess engines in sharp/tactical positions due to their insanely high nps advantage. SF managed to draw the reverse. Even LC has the same issue at least for now.
Pity Alekhine-Chatard is not normally reached due to Winawer.
7:35 "Ra3 you might think" I dont think anyone thought that :D
The french defense was my first black choice for a long time:/ I never like to see it crushed
Cheers, K
Well it's a bad opening lol.
@@JosephLachh it's not. Some variations are getting out of fashion but with those i use, the games get complicated and more importantly...equal.
Maybe king f8 to cover g7 instead of g6 creating holes in the dark squares near black king, but yes, the whole line looks fishy.
I would like to see the McCatheon with 4...Bb4 rather than 4...Be7
why not Nce4 immediately? this was my guess... right idea but wrong move order?? why move the bishop first?
Moral of this game: if you are black here, reject the gambit.
Bc4 is a triple exclam move.
7:31 Why do you think Alphazero should place the queen on A4? And I never heard Alphazero lost a single game to Stockfish.
It never lost any game when it played from beginning, it lost 6 games with it started playing from book openings.
Oh my god !!!! Oh!! What happened? ??? That is crazy!!
Fun comment - Cheers, K
i didn't like the french when i first saw it and wondered how anyone could feel comfortable with it. Then I looked into it and Stockfish finds e6 the best reply to e4. well, that's the french. so i tried it and found i liked it better than e5. maybe one day i'll learn how to play the sicilian. still, the french is solid!
very calm sea today at the dominican republic!
I would love to see more games w/ Stockfish as white!
"There is no way to get an advantage against the French that I know of."- IM John Watson. Think again. In fact, there is a wave of improvements and new ways to play against the French coming soon. French players will be devastated.
I recommend playback speed at 1.25
I went 1.5
Awesome game
Time:15:34,
knight D6? ;-)
Fire on the board! Amazing.
After ... axb5 Rxf7
Black should gladly accept a rook and knight for a queen. No need to go into hysterics. Neutralize white's attack.
Ng5 not a novelty I know that move from years ago check TCEC archive
Wow.. amazing game! So entertaining!
So, the classical variation of the French is losing for black. Guess what I'm never playing again!
Actually was thinking before you offered to pause the video about bishop e4. But after c6 couldnt find the right logic. Bishop c4! Right! Ok.
iirc playing to accept the pawn is losing.
brilliant game
Told ya.
A human would never accept the Chartard gambit
Fun comment - Cheers, K
@@kingscrusher Susan Polger at least in 2010 thought the Alekhine variation here was more interesting than good and white doesn't get enough compensation for the pawn. She shows a game where blacks plays pawn h5 to drive away the queen rather than h4 chosen here by alphazero.
I mean h6 chosen by aphazero to drive away or capture the knight.
Humans invented the Chartard gambit, you silly!
@@Prometheus4096 A human did invent it, but we need evidence that at least one human accepted the gambit. If everyone offered the gambit declined, then
Luke Chavhunduka
was correct.
i though alphazero never lost to stockfish only draw and wins?
That was the 100 game main match.
You need to improve explanation method
Typical human approach. Faced with an unbeatable chess engine they handicap it from the start by dumping a bad opening on it
and then rejoice in it losing to another unhandicapped engine.
Bang!!
Since when did alpha lose to sf? I thought A0 lost 0 games?! When was this played?! A0 and leela are trained in the game of chess. As soon as forced openings are employed it is not chess anymore it’s a variant of chess that they are not familiar with
John Davis the original report stated A0 lost 0 games!
John Davis thank you. This is news to me !!
John Davis I will check it out . I was under the understanding that A0 has not played again at all
A0 didn't lose any full games. The only games it lost contained book openings.
Andrew Smith right, A0 learned how to play chess. As soon as it’s forced to play specific openings it is no longer the same game it learned, it is nothing more than a variant of chess
But - the neural network will learn from this game!
Fun comment - Cheers, K
@@kingscrusher - Thanks KC - I really enjoy your chess videos.Well done.
Which "neural network"? Alpha zero does not use games against stockfish for training.
@@jellykin7161 - then I was misinformed. I thought it learnt from every game it played.
People believe whatever they want about alpha zero these days.
This is a horrible presentation given how good it could have been. I have watched hundreds/thousands (whatever) of chess analysis videos and this one is the worst out of anything/everything.
It's like Kingscrusher is so excited about what he seeing that he completely forgets about the viewer. Are you making this video for yourself to watch or is it for other people? (That's sarcasm.)
When you start, explain what is going on. For example, where did you get this game? Was Stockfish using it's opening book?
You talk so fast, machine gun firing fragments of words, it's ridiculous.
You're not differentiating between the moves of the engines and your own analysis, everything is blurred together.
This is the only AlphaZero analysis video that I had to stop and shut off because it is just so irritating.
Start over. Do it again a second time. Good luck.
More proof Alphazero is actually weaker than a proper Stockfish.
no, it was forced to play this variation. The french defense is shit.