Every GM for all of time: "You have to develop your minor pieces in the opening" Leela C0: *has every single back rank piece in its starting position at move 12 as white*
Its interesting how often neural networks create some sort of a zugzwang. Alpha also did this often to stockfish. Clamping it into a position where any move just makes stockfish's position worse.
Yay I am and excellent preparer for the next game So far excellent subscriber excellent rock lifter excellent manipulator of knight and an excellent winner when time is a trouble
@@DisplayLine6.13.9- I don't have enough chess expertise to judge everything said in the below video, but did find it very interesting. When Trading Bishop for Knight is Beneficial !!!!! - GM CHESS LESSONS ua-cam.com/video/RdUOyv3aO-A/v-deo.html
Both engine's evals were steadily growing throughout the game. The biggest boom for Leela (a point where the eval highly increases) was at the move 44.. Re8 after which Leela thought for 13 minutes and played 45. Rg3 (9:12 in the video) and increased eval from 2.98 to 4.61 pawns. After the monster move 67. Bd5 (14:08), which she played instantly, the eval stayed the same, but on her next move jumped from 8.09 to 11.46. The devs are now actually decreasing the evaluation function, because the values are too high, so the new formula will basically be the old one divided by 2.9.
agadmator, do you remeber one of the first games you played against Leela when she played Kg8 during the fried liver and you got the easy check mate? You beat the world champ!
Why couldn't you see LC0 eval? Just run the program, it's free. Though her eval is a bit different from what Stockfish processes, as Leela is based on win probability, not centipawn advantage count.
Probably not happening. AlphaZero was largely there to prove that neural networks have potential. Even then A0 fought Stockfish 8. Leela just won against SF 10. If A0 came back from the dead Leela would just yeet it back into its grave.
Alpha demolished stockfish 8 . They scored the same results against stockfish 9. And stockfish 10 is not so different than stockish 9. So Leela has no chance against alphzero. She cannot win a single game out of a thousand because without opening books alpha is almost unbeatable.
At 14:19 if bishop goes to c6 then it can't be captured by knight as pawn captures with check and if king captures the pawn then knight will deliver a check winning the rook at f8 Just an idea
#suggestion This is a good time to show the first game of the first chess program. The program was written by Alan Turing before the first computers could possibly run it. He had to do the calculation on his own on paper. The game was played between "The Paper Machine" (Alan Turing) and Alick Glennie in 1952. Here is a nice article about this machine and the game: en.chessbase.com/post/reconstructing-turing-s-paper-machine
I watched all those leela games on you channel. I'm quite bad at chess and mostly interested in AI's more than chess, but from my perspective I feel like a major difference between stockfish and AI's like leela or alphazero is that stockfish doesn't have a gameplan. If it feels comfortable in it's position and see no real opportunity to improve it, it will be content shuffling it's pieces a bit without achieving much with them. Of course that make for a very solid wall, impossible to break for a human and it still punishes every mistake. But what happens when you don't make mistakes and have a gameplan is that while the engine is shuffling somewhat aimlessly (well with the aim of keeping it's good position), the AI will be actively looking to create opportunities to win, and despite both boards being somewhat equal it'll prepare a move for a later and then try to enable that move. I feel that's the real différence and the reason why those AI's are so strong.Then again I'm pretty bad at chess so I might be wrong, and of course I'm antropomorphizing a bit cause it's not really planning, it's pattern recognition, but it works functionnaly the same.
At Dictionary.com, the second meaning is : 2.imitation; simulated; sham: Artificially does not have to mean man-made. Simulated castling move(s) is artificial castling no matter who/what does the moving.
I like how the moves are so human-like. Not like some games from TCEC where two engines just make _number_ of moves with 2-3 pieces back and forth in the endgame. Here, it every move that was played, was relatable
Thank you so much CM Agadmator for making these videos. I started watching you a month ago and I have improved 200 rating points since then this game is awesome too, so nice to watch
I'm wondering if it ever will be possible for chess engine to calculate all possiblities of the game, and we will see from start mate in 100 moves or something like that, or it will be always a draw.
Solving chess is pretty much impossible for now and will be for a long time. English draughts have been solved (the game is a draw) and it required dozens of computers over a couple decades to accomplish this. The game-tree complexity of chess is 10^92 times larger than that of draughts for comparison.
A year after the ceration of Leela, it is nice to see that Leela is now the champion. Before, Leela can’t even get into the top 5. It is very interesting to see her play against Alpha Zero.
A0 is weaker, Leela is much more complicated and uses modern deep learning technologies like squeeze-and-excitation, varying learning rate etc. which a0 doesn't.
Glad they have the 'learn from your mistakes feature on lichess now' But seriously - what really sticks out from all these AI games is how they evaluate the power of pawns. They seem to know exactly when to sacrifice them for a powerful attack and how to play a long term defensive strategy to win in the end game. Just amazing stuff. I'm so proud of the leela developers for going through with the experiment after google decided not to continue... It would be funny if the people at deepmind 'accidentally' left the alpha zero algorithm continue to run all this time and it came back with a vengeance to crush leela :p
the fact that leela is crying while staring me right in the eye with green face on her thumbnail, makes be relaxed that the ai revolution will work out fine
At 14:00, N. Be6+ N... Nxe6 N+1 e6+ N+1 ... N+2 Ne7+ N+2 ... K d7 N+3 Rf8 N+3' ... Ke7 N+4 Ra8. Black left with a knight and a few soon to be lost pawns - White with a rook and two connected King side pawns.
In this game SF slowly realized it was loosing, it was around +1 from move 25, but did not reach +2 until move 50, +3 by move 67 and then it quickly rose to +10 by move 78. The "TCEC win rule" which ends the match is that both engines have to be +10 or more (or below -10) for 8 ply (4 moves).
Theoretically, at any given point the outcome of the chess party is decided but because people and computers cannot look far ahead, they make suboptimal moves which creates opportunities for the opponent that were not available earlier. This is what makes chess and games like that, like Go, so interesting. It is very different from games of chance like Bridge. In Bridge, you try to reduce as much as possible the randomity that chance and luck brings. In chess, you try to increase the randomity as much as possible in order to increase your chances and luck.
Solid analysis of this game. Many alternatives discussed. Nice. I think Leela would do OK against Alpha Zero. Both would have trained about the same amount by that time and assuming that both were running out the same hardware. I don't see why Alpha necessarily has the advantage unless it is running on much better hardware.
Agad! Analyse some of your games soon! I wanna continue the Capablanca saga and of course there are going to be a lot of chess games to see for the next month, but I wanna see some more of your games. You're a Candidate Master after all
11:40 Leela could play Rgxf5 Rxf5 Rxf5 Rxf5 and then Ke7+ picking up the rook Edit: And Leela would have two passed pawns on the king side which should lead to victory
because of it's very strong hardware I think. I have heard about that alpha mastered chess in about 4 hours of playing with itself. it played maybe hundreds of matches at the same time and they probably lasted like few seconds each.
How many more principles will these engines break. After move 12 all of Leila’s pieces are in their starting positions, I know it’s a wired opening but this keeps happening and I’m starting to question all chess theory
I think because the AI/engines have 'discovered' that the combined pawn structure is the most important thing to focus on thus they are playing the pieces to support that.
Turns out i am an excellent finder of these super moves, though i still have no idea how i find them. Bd5 looks natural to me, but can’t see the “why”. How can i get better at this?
I thought that maybe a plan for Stockfish was to relocate the knight from a5 to d4 much earlier. On a5 it does nothing but on d4 it is in the center and controls key squares as well as defending f5.
Congrats on becoming a CM Agad! Well deserved :)
When did he become a cm bro
@@14b3am check his ig
A GM?
@@RikardoAHP CM is Candidate Master
Every GM for all of time: "You have to develop your minor pieces in the opening"
Leela C0: *has every single back rank piece in its starting position at move 12 as white*
To be fair those positions were played by humans and set up, I don't remember if up to 12 but 10 for sure.
@@Saludin2 app pieces are on the back rank at 10 and 12 moves, move 11 is the queen check followed by putting the queen back
It’s ugly , but it’s a move - Agadmator
he keeps the things neat and short.
When you ask to marry your girl
1:46 and from this point we have a completely new game so stockfish definitely came prepared
www.d-e-f-i-n-i-t-e-l-y.com/
What does the comment even mean lol
@@owurakuboadu the comment above before edit said defiantly
@@rmcoooo yah I edited it thanks for pointing the typo
@@rmcoooo nice idea for grabbing some money from Google tho !!
Its interesting how often neural networks create some sort of a zugzwang. Alpha also did this often to stockfish. Clamping it into a position where any move just makes stockfish's position worse.
Yay I am and excellent preparer for the next game
So far
excellent subscriber
excellent rock lifter
excellent manipulator of knight
and an excellent winner when time is a trouble
I am one of those that just want to see the show
@@MattMoody5225 You should rather enjoy it though
Congratulations, you are also an excellent commentator.
Show us the game in which you became a CM, Please!!!
It wasn't a single game. It was a series of games.
@@kingofchase2539 the one that he found the most interesting then
When this happen? Someone send me his page on the fide web site?
@@Devp278 yeah I bet there must be a couple. He only lost 1 in the tournament. Maybe he could show us that.
Maybe he made some ugly mistakes n now feeling shy. Bring it on, Agad.
What did I learn from watching this game. You never want your knights to guard each other. Maybe you advanced players knew this, but I did not.
Yeah but what if one of those knights is a monster knight ?
@@DisplayLine6.13.9- I don't have enough chess expertise to judge everything said in the below video, but did find it very interesting.
When Trading Bishop for Knight is Beneficial !!!!! - GM CHESS LESSONS
ua-cam.com/video/RdUOyv3aO-A/v-deo.html
Both engine's evals were steadily growing throughout the game. The biggest boom for Leela (a point where the eval highly increases) was at the move 44.. Re8 after which Leela thought for 13 minutes and played 45. Rg3 (9:12 in the video) and increased eval from 2.98 to 4.61 pawns. After the monster move 67. Bd5 (14:08), which she played instantly, the eval stayed the same, but on her next move jumped from 8.09 to 11.46.
The devs are now actually decreasing the evaluation function, because the values are too high, so the new formula will basically be the old one divided by 2.9.
Mnd
That is right
agadmator, do you remeber one of the first games you played against Leela when she played Kg8 during the fried liver and you got the easy check mate? You beat the world champ!
Good times :)
Hahaha sure he did
He beat her when she was an infant. I smile as I picture an infant AI sitting at a chessboard opposite agadmator.
It's like playing against a chess prodigy - beat them while you still can!
13:45 -- "Feel free to pause the video..."
Yesss!! First time I guessed it right.
I hope I have my day too. . .
I wish we can look at LC0 eval on a game instead of stockfish
You can if you use your GPU to run Leela
You can on the TCEC website.
ua-cam.com/video/vGcjqwUniM4/v-deo.html
Why couldn't you see LC0 eval? Just run the program, it's free. Though her eval is a bit different from what Stockfish processes, as Leela is based on win probability, not centipawn advantage count.
You can see the win percentage and eval for every single move on the TCEC website.
Agadmator you really are the best youtuber, thank you for everything
We need alpha zero against Lela games
So we can watch the rocks destroy each other
Probably not happening. AlphaZero was largely there to prove that neural networks have potential. Even then A0 fought Stockfish 8. Leela just won against SF 10. If A0 came back from the dead Leela would just yeet it back into its grave.
@@A_Swarm_of_Waspcrabs Why did Agadmator say that AO would probably win all the games against Leela at the end then?
Alpha demolished stockfish 8 . They scored the same results against stockfish 9. And stockfish 10 is not so different than stockish 9. So Leela has no chance against alphzero. She cannot win a single game out of a thousand because without opening books alpha is almost unbeatable.
Knight on a5 : What is my purpose ?
Stockfish : You attack pawn
Knight on a5 : Oh my god...
9:59
Agadmator about Leela's position: "It's just better"
StockDog: **Leaves the room**
Best quick chess analysis ever, i'm so impressed at how good agadmator has gotten in the 2+ years I've been following him
Stochfish went for a Temporary pawn sacrifice. Literally did not regain the pawn.
Pawns are equal after 9:20
Stochfish? Is that his Hungarian brother?
I thought a.i.'s always castle....
Artificially? ;D
MeEp ChaoTic a very good one
Nice one sir
'.' did laugh at bad joke
Outstanding move
Show us how you got your cm title :)
He bought it
At 14:19 if bishop goes to c6 then it can't be captured by knight as pawn captures with check and if king captures the pawn then knight will deliver a check winning the rook at f8
Just an idea
#suggestion
This is a good time to show the first game of the first chess program.
The program was written by Alan Turing before the first computers could possibly run it.
He had to do the calculation on his own on paper.
The game was played between "The Paper Machine" (Alan Turing) and Alick Glennie in 1952.
Here is a nice article about this machine and the game:
en.chessbase.com/post/reconstructing-turing-s-paper-machine
I watched all those leela games on you channel. I'm quite bad at chess and mostly interested in AI's more than chess, but from my perspective I feel like a major difference between stockfish and AI's like leela or alphazero is that stockfish doesn't have a gameplan. If it feels comfortable in it's position and see no real opportunity to improve it, it will be content shuffling it's pieces a bit without achieving much with them. Of course that make for a very solid wall, impossible to break for a human and it still punishes every mistake.
But what happens when you don't make mistakes and have a gameplan is that while the engine is shuffling somewhat aimlessly (well with the aim of keeping it's good position), the AI will be actively looking to create opportunities to win, and despite both boards being somewhat equal it'll prepare a move for a later and then try to enable that move.
I feel that's the real différence and the reason why those AI's are so strong.Then again I'm pretty bad at chess so I might be wrong, and of course I'm antropomorphizing a bit cause it's not really planning, it's pattern recognition, but it works functionnaly the same.
What a nice game of Evan's Gambit :)
Great video, really enjoy these.
And good job UA-cam, actually showing me an ad that was relevant to the video.
2:27 can stockfish really "castle artificially", cause like artificially means man-made....
At Dictionary.com, the second meaning is : 2.imitation; simulated; sham:
Artificially does not have to mean man-made.
Simulated castling move(s) is artificial castling no matter who/what does the moving.
They both agreed that stockfigh was losing :D Really like such expresion :D
3:00 what would happen if stockfish captured on b2 instead of moving the knight ?
Also interested in this.
I guess there would be a ton of activity for Lela with black´s king in the center
I like how the moves are so human-like. Not like some games from TCEC where two engines just make _number_ of moves with 2-3 pieces back and forth in the endgame.
Here, it every move that was played, was relatable
Leela will definitely compete in future TCEC seasons! She will get stronger too, the net that won this sufi hadn't finished its training run.
Thank you so much CM Agadmator for making these videos.
I started watching you a month ago and I have improved 200 rating points since then
this game is awesome too, so nice to watch
I think at 15:13 the king c7 move was to bring the knight back into the game so it could play knight A5 then to B7, a previously unguarded square
I'm wondering if it ever will be possible for chess engine to calculate all possiblities of the game, and we will see from start mate in 100 moves or something like that, or it will be always a draw.
Solving chess is pretty much impossible for now and will be for a long time. English draughts have been solved (the game is a draw) and it required dozens of computers over a couple decades to accomplish this. The game-tree complexity of chess is 10^92 times larger than that of draughts for comparison.
3:03
Why not capture on b2? I get that the queen would have to quickly run away and white would get a development lead, but a pawn is a pawn.
That's exactly what I came to the comments to ask. I'm also curious.
19:39 "Superior Chess Entity" drives fear to my bones!
16:40 B-f7 also looks strong, seems to trap the Rook. If NxB, then white N-f5 forces the Rook back, then PxN looks like a quick win.
The evaluation goes over 9000 for Leela :D
Did he mention that all moves until after 7.Bc1 were part of the opening book?
In other words, the games started from the position after 7.Bc1
The opening moves are given to the engines yes and they both play white and black each in each opening.
My doctor has recommended that I avoid artificial castling.
My wife has artificial castles and I absolutely love them.
What if (3:04) bishop captures b2?
If bishop captures b2, white bishop captures and queen captures. Black won a pawn or am I missing something?
Same question
Same question. Can anyone explain?
A year after the ceration of Leela, it is nice to see that Leela is now the champion. Before, Leela can’t even get into the top 5. It is very interesting to see her play against Alpha Zero.
At this point I think is not that obvious that Alpha is still stronger than Leela
I think it would depend on the hardware, but unfortunately the chances of a A0 vs Leela match is near 0.
A0 is weaker, Leela is much more complicated and uses modern deep learning technologies like squeeze-and-excitation, varying learning rate etc. which a0 doesn't.
Glad they have the 'learn from your mistakes feature on lichess now' But seriously - what really sticks out from all these AI games is how they evaluate the power of pawns. They seem to know exactly when to sacrifice them for a powerful attack and how to play a long term defensive strategy to win in the end game. Just amazing stuff. I'm so proud of the leela developers for going through with the experiment after google decided not to continue... It would be funny if the people at deepmind 'accidentally' left the alpha zero algorithm continue to run all this time and it came back with a vengeance to crush leela :p
yay I found the crazy super move!
I’m proud to say I did a super punch to the sky to celebrate :P
after watching like 60 vids finally I'm an excellent at something
Same here. I am an excellent finder of super-moves
I'm an excellent preparer😂
18:13 What's wrong with grabbing c4 pawn immediately?
"The superior chess entity"
Now that sounds ominous
the fact that leela is crying while staring me right in the eye with green face on her thumbnail, makes be relaxed that the ai revolution will work out fine
I usually don’t watch computer chess but damn that was pleasing to watch. Great video man
Hey what chess gui program do you use? with what settings?
Thank you for pronouncing my name perfectly!
May I suggest "Who Let the Pawns Out?" - Kramnik v Kasparov
#suggestion
At 14:00, N. Be6+ N... Nxe6 N+1 e6+ N+1 ... N+2 Ne7+ N+2 ... K d7 N+3 Rf8 N+3' ... Ke7 N+4 Ra8. Black left with a knight and a few soon to be lost pawns - White with a rook and two connected King side pawns.
In this game SF slowly realized it was loosing, it was around +1 from move 25, but did not reach +2 until move 50, +3 by move 67 and then it quickly rose to +10 by move 78.
The "TCEC win rule" which ends the match is that both engines have to be +10 or more (or below -10) for 8 ply (4 moves).
Leela is definitely stronger than the first version of Alpha Zero, there was a simulation not long ago on TCEC bonus server proving that.
link?
Can someone pls make a compilation of Agad saying "hello everyone"😂
What's the software (chess engine ) u use in vedeos ?
The awesomeness of the analysis overwhelmed the dog in the background.... So finally he left his mark🤣😆
best chess channel in the world
Theoretically, at any given point the outcome of the chess party is decided but because people and computers cannot look far ahead, they make suboptimal moves which creates opportunities for the opponent that were not available earlier. This is what makes chess and games like that, like Go, so interesting. It is very different from games of chance like Bridge. In Bridge, you try to reduce as much as possible the randomity that chance and luck brings. In chess, you try to increase the randomity as much as possible in order to increase your chances and luck.
Time for agadmator saga
Solid analysis of this game. Many alternatives discussed. Nice.
I think Leela would do OK against Alpha Zero.
Both would have trained about the same amount by that time and assuming that both were running out the same hardware.
I don't see why Alpha necessarily has the advantage unless it is running on much better hardware.
But they havent trained the same... alpha zero has trained significantly more than leela
why did alphazero not participate in this tournament?
Here's an idea. Could you show every players chess rating? I think this small thing would help me know what rating are the players :3
Leela's elo was 3650 during superfinal :)
Are there any games between Leela and Alpha Zero? I would love to see that. I noticed that Alpha wasn't in this or the last tournament.
Agad! Analyse some of your games soon! I wanna continue the Capablanca saga and of course there are going to be a lot of chess games to see for the next month, but I wanna see some more of your games. You're a Candidate Master after all
18:14 why knight not capturing a4 pawn?
You can see the evaluations in the tcec archive: cd.tcecbeta.club/archive.html?season=15&div=sf&game=62
11:40 Leela could play Rgxf5 Rxf5 Rxf5 Rxf5 and then Ke7+ picking up the rook
Edit: And Leela would have two passed pawns on the king side which should lead to victory
Then you lose the c4 pawn next move and black has 2 connected passed pawns in the center
what abou alpha zero? did he play the tournament?
Did you show the season 14 superfinal in which stock fish defeated leela?
Well explained sir.. I really appreciate the way explaining the game..
Captures, captures, captures, captures, captures..... 😛
Positional chess is back!
Karpov revival. But not only - AI's also play very aggressive attacking with lots of sacrifices, if the position allows it.
Sorry if the question is stupid, but why doesnt Alpha Zero compete? And why is it so much superior to the other engines?
Alpha Zero retired from chess (and Go and Shogi).
because of it's very strong hardware I think. I have heard about that alpha mastered chess in about 4 hours of playing with itself. it played maybe hundreds of matches at the same time and they probably lasted like few seconds each.
@@plaguedoc1403 So, it diddled itseself? The Big M word here.
Bd4 ... and the piece stayed put! Your analysis was on the mark; very precise continuation from Leela.
d5 of course. I would play c6# big mistake.
I have a question : How long does it take the 1 enginges to complete a game? Does it happen instantly?
Extremely enjoyable game especially with your explanation guiding too.
Can someone please tell me, if you know, where can I download Leela's profile photo from? I searched all over Google and couldn't find the same photo
Agad's speed increases when resignation is near.
Funny how he said at 2:30 "artificially" referring to Leela
Superb explanations on lines Agad I just love your videos
Nice quote above the board 😂
which version of stockfish is this?.... stockfish 10?
@Fanne Aeris 14-7
10 but dev version
Did Alpha Zero play in this competition?
I am addicted to agadmator chess videos and result is that i can't break my loosing streak...
Great video! Thank you for your analysis.
Can someone tell me what's wrong with 74. Be4, tripling protection on the g6 pawn?
How many more principles will these engines break. After move 12 all of Leila’s pieces are in their starting positions, I know it’s a wired opening but this keeps happening and I’m starting to question all chess theory
I think because the AI/engines have 'discovered' that the combined pawn structure is the most important thing to focus on thus they are playing the pieces to support that.
would you ever do viewer games? I would like to see how good other people are and see how players that arent gms do, and idk maybe it would help me
Aga, about how many over the board rated games have you played?
Hey Agad can you do some videos of stockfish or leela vs weaker engines (e.g. play magnus or win7 chess)?
Well now there’s only one thing left to see. Leela vs Alpha
Has Leela played Alpha Zero? That would be amazing to watch
Alpha Zero has not entered any championships. The only game records we have are private games against Stockfish.
Who is better? Leela or Alpha zero?
Whenever in doubt, just push f5 :)
1. e4 f5
What's the time control?? I wish you would say what it is.
Google TCEC
That was intense.
Turns out i am an excellent finder of these super moves, though i still have no idea how i find them.
Bd5 looks natural to me, but can’t see the “why”. How can i get better at this?
Is this stockfish 10?
I thought that maybe a plan for Stockfish was to relocate the knight from a5 to d4 much earlier. On a5 it does nothing but on d4 it is in the center and controls key squares as well as defending f5.
Apparently SF didn't have time to do that earlier.
I hope we’ll see match AlphaZero against Leela Chess Zero 😍