The impressive factor is Stockfish evaluates 100,000,000 positions but Leela only evaluates 100,000. In terms of position evaluation, Leela compensated for a loss of 7 to 8 moves ahead by superior position evaluation : equivalent to correctly evaluating its position 10 moves before Stockfish. There is no doubt Leela with sufficient computation power to evaluate 100,000,000 positions would crush Stockfish : Alpha Zero with the Google network would crush such a Leela.
Try blitz on lichess (and probably any other chess server). Once you have strong advantage in endgame (like Queen+) opponent starts giving up material trying to get a stalemate. Happened to me once or twice in about 14,5k games that I played , opponent somehow managed to get a stalemate due to me being low on time and making a wrong move against sole king.
No is not, it is counter intuitive only for people with a closed mind and limited culture. 🤷🏿♀️ "He who knows when he can fight and when he cannot, will be victorious." (Chapter III: Attack by Stratagem) Art of War. Sun Tzu
There was a game with Samuel Reshevsky vs (I believe) Edward Lasker, where Lasker had a dominant position (pawn about to promote, mate shortly following) and Reshevsky reached across the board - and Lasker thought he was going to shake hands for the loss - when Reshevsky picked up his queen and checked the white king with no choice but to capture. Then, he proceed to slide his rook along the 2nd rank, check. if the king captured, it was stalemate with no legal moves for black and if the rook was not captured, it would be a draw by perpetual check. Similar situation to this game. The lesson learned: even in a hopeless position - a draw might be swindled.
Fabulous. I pine for the days when Leela was the only neural network in town and Stockfish was just 'all the games humans have ever played'. Leela woke the world up to first principles then and was allowed to play stunning innovative chess.
Actually, Leela has defeated Stockfish twice in the TCEC Grand Championship, three times to take the TCEC Cup, and the last two times to win the TCEC Swiss. So so it's not like the Fish has had no competition since 2019.
Generally speaking stockfish does better against weaker engines, so it wins most of the round robin tournaments, but yeah, in head-to-head leela does quite well.
And actually, Stockfish has won every event on CCC (where this game is from) since 2020. And actually, Leela hasn't finished better than 4th in the last three bullet events.
Great game and great descriptions of the difference in engines. One minor improvement: Leela doesn't use NNUE. It actually uses "inefficiently updatable" neural networks in the sense that one needs to upload data to the GPU to perform computations. Stockfish uses CPU exclusively, that's why it needs NNUE: CPU cannot evaluate the deep neural networks Leela uses, as GPU does.
@@jx14aby what an awful thing to say. the founder of AlphaZero recently won the noble prize because the work he did developed the nueral nets for AlphaZero were able to be applied to predict protein folding with AlphaFold. It saves researchers thousands of manhours and allows for massive advancements in the medical industry. If he didn't "waste his time working on computer chess" the world would be deprived of of the greatest achievements in recent history.
Well, Leela on CPU can still beat grandmasters easily. Just will lose a few hundred ELO from having the nodes searchable being reduced by about 2 orders of magnitude.
Not many engine like moves in this game surprisingly. Maybe the black Q to a1, maybe... A lot of really bad choices in this game, any chess player can see right away. More like a composed game really to come to this draw. Strange game. Engine games usually are strange, but in an incomprehensible way. There were blunders in this game too, which is unlikely for engines. Ok, there's blunders in every game you can argue, even in eng games, but when human chess players can see them... I call fraud in this game.
@@dariogreggio7981 Leela is the home brew version of Google Alpha Zero. At no point has it had the computational brute force available to it, to match Alpha Zero.
This game shocked me. Leela was so forward thinking that it saw winning was not possible and orchestrated this outcome a dozen moves in advance and stockfish seemingly was helpless and unaware due to its own single-minded ruthlessness to be aware of Leelas strategy.
These algorithms seem to have personalities. Anybody know why people associate Leela as a she/her but stockfish has no associated gender? Is neural transgender a thing?
I once gave rook odds to my friend, but he played well, got a passer and when position became unwinnable I sacked my other rook and the queen for the stalemate. He was looking at the board for another 20 minutes. It could be his first win in years, but I wiggled out. Still the best game of my life.
Antonio’s Brother in law also has a Chess channel Jozarov Chess he shows many Stockfish and engine games as well as the top GMs and does a great job as well
@@michaelmassaro4375Yes, Jozarov is the best (:) at commentating engine games. He sometimes goes through like 20 different variations/sidelines but somehow never loses me throughout. Great to learn from his presentation of the ideas in the game
Pure poetry! It taught us that if we're losing and we can sacrifice everything getting a stalemate, this is one of the most "simple" ways to get a draw💪
So I am a SUPER new chess player, and it will probably sound very naive, but is the game a draw because the black pawns cannot move and ANY move the king makes is checkmate? I didn't even know this was a rule, if I was playing my uncle I would assume this was checkmate because the king would HAVE to move into a death position... I guess that's not the case... Fascinating!
Yeah this is one of the more unfair seeming so-close-yet-so-far type situations in chess. If the king is not currently under check, and cannot move to any square where it isn't at check, while also having no other possible legal moves, it is called a stalemate, and considered a draw. It's actually one of the things top players are trained to be wary of in end games where they have a massive advantage.
Honestly this is one of the greatest games ive ever seen😢. We see miracle wins all the time but you almost never see crazy stalemate draws. Unbelievable 😍
Leela remembered us a lesson often forgotten: in every position, the key is the king. You should always start evaluating a position after understanding if your king is good or not. After all, the king is a tactical weakness. In this game, white sacrificed for the whole game the safety of his king in order to attack, while Leela tried to play solidly without giving too much. Stock fish only had a mating treat in the last move he played, and that’s exactly where Leela showed us the importance of checking your king position.
Eventhough its not impossible but when i saw the title i thought Stockfish lost with white pieces and i really amazed. But anyway draw is acceptable. They have huge depth of calculation and seeing these lines are possible. Thanks for this great game Agad.😊❤
@@hosseinmohammadi8776 I remb I saw a game where stockfish lost as white when the opening evaluation from stockfish was 0.0. So it not necessarily disadvantage position
@ndnd7614 wow, interesting. Eventhough a bit unbelievable.😊 Maybe a blitz game or I don't know anything can happen but forexample stockfish 15 with that strong depth Plus nnue , for me it's strange. But I have to know about it more. Thanks for your statement. It made me to go and search more about it.😊🤚
There was also an amazing game from another RUclipsr called ASMR Chess where he bought a super old chess machine called Mephisto (like from the 1980s old) and put it against Stockfish. The wise old Mephisto also held Stockfish to a draw! Granted, they started from a normal, fresh game instead of pre-programming to start in a Caro-Khan, but it's still pretty impressive for such an old computer.
What a video! TLDR, As a data scientist, what happened here (for people that doesnt know too much math), suppose that you are going from a city A to B with a car and think thats the fastest way to do it is take C1 runway (100 km), but sometimes it gets lots of traffic using time to travel about 200 km, and C2 (150 km) doenst get traffic. We can decide into C2, but stockfish actually doesnt. It wasnt a lc0 art, since stockfish knew the drawish lines. But this didnt make this video worse at all. If my hypothesis is correct, this game shows perfectly the psychological factor in chess. From the move 38 Nxb5, the game is completely drawn but many people couldnt find the line after being positionally crushed for about 40 moves 6:55 I dont know how lc0 evaluates (if engines also generates a log from those matches...) but this can be a "bluff". Lets say do you have 5 suboptimal moves, maybe do the "best" supoptimal leads to a easily eval (lets say that best suboptimal leads to material loss in 5 moves and others in 10), so its better to do something thats get you worst, but it takes more moves (and more chance of errors). 6:40 is the same concept, how 3x0 is better than 3x1? it isnt. the difference is, stockfish found the trick on 6:40. I cannot prove it, but can give some decent imagination. After 28. Qc5 stockfish (depth 33) evaluates 5 moves (ordered) for leela (b6, Rd5, h6, Rfe8, Qc8), scoring from 1.37 to 2.19, with the next move for stockfish (ordered) (c6 a6 O-O-O O-O-O O-O-O) When leela executes 28... b6, the 5 lines are (c6 1.42 cxb6 0.11 Rg3 -1.00 Kf2 -1.16 a6 -1.21). If 28... Rd5 was played, all lines are +1 for white. So 28...b6 and 28...Rd7 gives the same evaluation locally, but b6 allows some "chance" of blundering. Finally, lc0 pratically loses by brute force, but the little piece of "human playing" saved lc0, since it doesnt not eval how to win, but how to not lose. This could be a good test for stockfish developers: how to make the engine, when completely winning, avoid chances of not winning by this kind of situation by, as an example, giving some material. Again, cant prove it, but as an example: Even at lower depth (33) the drawish lines starts to appear after 38 Nxb5 Qb6, another line is 39 Nd6 Bxd6 40 Rxd6 Rxd6 41 Bxd6 Bxd6 42 Bc5 Qb5 53 Kf2 Ba8. Completely drawn (all evaluations for stockfish in this line are play or get worse and the result is always +0). Instead. 38 c7 didnt raised any drawn line (depth 33), always giving +2 for white. This problem can be saw as finding a local optimum that wasnt the global optimum. Without deeper analysis (this took something like 15 minutes for me) stockfish blundered with 38 Nxb5 (even being the best move), leading a superior position into a drawn state. Thats why some algorithms like simulated annealing, genetic optimization etc are essential.
I really enjoy watching these strong engines play! I can just enjoy the great chess without having to care who wins. And extravagantly good chess this was!
@@kingStribor 3 is not 1. And 3^2 is not 1. So they are not playing the exact same opening. Instead they variate between different openings. Even if between only a few.
it is really amazing how much poison there is in the position at move 41-42 even with such a dominant position a check mate is possible with a simple mistake !!
@@507RiverRockyeah but I wonder, could Stockfish have spotted this earlier and leveraged it’s dominant position and when did Stockfish started computing those absolute GOAT moves
Not necessarily. Stockfish could not have understood what was going to happen when it played Nxb5, because it had alternative moves which preserved its advantage. But Leela might not have understood what was going on at that point either; it might have just kept playing the only move that didn't lose immediately, without knowing that it could draw.
@@JohnDoe-ti2npyeah that’s possible I really wonder that’s why I would be really curious if there is anywhere we could see the rating each engine give at the position on the moves to see when each see that
Stockfish should’ve used its Spider 🕷️ Sense hey Leela giving me pawns for free ?? What up with that but nope super nice moves from Leela disposing of its property
Hey Agad I love all your videos on going over chess engine games, very educational. I was wondering if you've heard of an engine called Patricia, I feel that could make for some good content and games to go over
I think it's the first time I see a game ended in a draw in this channel... I very much liked it and I think it would be great to show more games ended like this, sure there are a lot that have been interesting between humans too (greetings from Ecuador)
The chess computer on my phone does this to me all the time. By the time I realize its chasing me down with a piece I simply can't capture, it's already too late. For humans we don't intuitively look for this, but for a silicon brain its just yet another permutation that's part of its calculation methodology.
I've sacced all my pieces to make forced draws a few times (though not 4 of them as here?), but I didn't celebrate it, because I was usually uncertain as to whether doing so was best.
This reminds me of my first time playing chess. We were in High School and I just learned how to play chess. I am always losing to my friends but I accidentally discovered stalement. They were amused by that time because they did not expected a newbie to pull off a stalemate. Since that day I think they fear me for playing sacrifices because I have been known for stalemates. Good old times😅
Hi. It may be because Bd7xa4 sets up a self-pin (the bishop is now pinned against their own Queen on a7 and White has a rook on a1). If the Queen moves, the bishop is now en-prise, so Black effectively has two pieces out of action until the pin can be resolved.
@@Timepasser_me Yes, ..Ra8 can be played, but the rook is behind the Queen, so it is still pinned. If things were the other way round, i.e. Queen on a8 and Black could play ..Ra7, then that would work as it would break the pin and still defend the bishop on a4. If the bishop on a4 was then moved, Rxa7 is met by Qxa7.
Leela chess zero is the alternative of alpha zero made by the same programmers so that just means if stockfish is better than leela its also better than alpha zero
Hi agadmator! Quick question about Chess engines: Can engines know if the game is a chill training one vs a tournament one? I would assume a good strategy might be to practice uncommon/weird moves that the other Algo might not have checked. Assuming algos didn't to full breath searches and they trim certain options, playing such moves might being the game to a place where the other Algo might be at a disadvantage. Thoughts?
- Agad: " Pause the video and figure out what a 3600 ELO engine played in this position "
- Me, a 1100 ELO player: " Yes."
me a non rated newbie: "cecshs"
"...while I give you a few seconds..."
but i found the move, even tho im a casual player
I thought rook to f8. Thats why I’m 912 blitz and 1300 rapid.
😂me too!
bro calculated stalemate in 22
😃
What matters is that:
1) Leela saw the draw 3 moves before Stockfish
2) Stockfish was already at +4.0 hence the "shock"
Shockfish
Imagining stockfish mouth suddenly opening like an overacting karen getting talked back at
The impressive factor is Stockfish evaluates 100,000,000 positions but Leela only evaluates 100,000. In terms of position evaluation, Leela compensated for a loss of 7 to 8 moves ahead by superior position evaluation : equivalent to correctly evaluating its position 10 moves before Stockfish.
There is no doubt Leela with sufficient computation power to evaluate 100,000,000 positions would crush Stockfish : Alpha Zero with the Google network would crush such a Leela.
Leela: Oh no my bishop.
Leela Finegold?
I see what you did there..
Leela Rosen?
hahahahahahaha
@@jeffreywilliams3421if it was Leela finegold it must've said I am Grand master Leela and you are not boom boom 😂😂
That is actually the craziest draw I've ever seen
Fr💀
Same here never seen something like this, truly brilliant
i have seen some worse lol its crazy
I thought Leela would win, Did not expected a crazy stalemate with sacrificng 3 pieces.
@@Kitarp064 pieces
I never trust stockfish. Especially when it says I made 22 blunders in one game
😂😂
😂😂
Surely you've made at least 30
Rookie numbers. Gotta pump those numbers up!
agreed!
who is stockfish to tell me I made a blunder in move 2?
Incredible! Forcing a stalemate by sacrificing all the remaining material is so counter-intuitive for the human mind.
Really? Thats one of the first things that comes to my mind to achieve stalemate
Try blitz on lichess (and probably any other chess server). Once you have strong advantage in endgame (like Queen+) opponent starts giving up material trying to get a stalemate. Happened to me once or twice in about 14,5k games that I played , opponent somehow managed to get a stalemate due to me being low on time and making a wrong move against sole king.
@@logangustavsonnot so much material
it was quite counter-intuitive for SF as well tho :D
No is not, it is counter intuitive only for people with a closed mind and limited culture. 🤷🏿♀️
"He who knows when he can fight and when he cannot, will be victorious."
(Chapter III: Attack by Stratagem) Art of War. Sun Tzu
There was a game with Samuel Reshevsky vs (I believe) Edward Lasker, where Lasker had a dominant position (pawn about to promote, mate shortly following) and Reshevsky reached across the board - and Lasker thought he was going to shake hands for the loss - when Reshevsky picked up his queen and checked the white king with no choice but to capture. Then, he proceed to slide his rook along the 2nd rank, check. if the king captured, it was stalemate with no legal moves for black and if the rook was not captured, it would be a draw by perpetual check. Similar situation to this game. The lesson learned: even in a hopeless position - a draw might be swindled.
The shit eating grin on Antonio’s face when he gives us a couple seconds to figure it out is pure gold
His eyes literally gleam with schadenfreude🤩
Lol 😂
🤣🤣🤣
Fabulous. I pine for the days when Leela was the only neural network in town and Stockfish was just 'all the games humans have ever played'. Leela woke the world up to first principles then and was allowed to play stunning innovative chess.
engine playing style is very solid, will not give the opponent space, continues to press and always counters every move
Well, it’s an engine
Inaccuracies are like blunders to them.
Inaccuracies? Nah even excellent moves are blunder for them @@kokoro212
no even if your move are all good move it was not enough to beat engine.@@kokoro212
But they can't counter my illegal moves to win
Actually, Leela has defeated Stockfish twice in the TCEC Grand Championship, three times to take the TCEC Cup, and the last two times to win the TCEC Swiss. So so it's not like the Fish has had no competition since 2019.
Generally speaking stockfish does better against weaker engines, so it wins most of the round robin tournaments, but yeah, in head-to-head leela does quite well.
what about alphazero
@@andrewli1455 alphazero is discontinued
And actually, Stockfish has won every event on CCC (where this game is from) since 2020.
And actually, Leela hasn't finished better than 4th in the last three bullet events.
Thanks for bringing more computer games
Any favorite computer games you reccomend?
Great game and great descriptions of the difference in engines. One minor improvement: Leela doesn't use NNUE. It actually uses "inefficiently updatable" neural networks in the sense that one needs to upload data to the GPU to perform computations. Stockfish uses CPU exclusively, that's why it needs NNUE: CPU cannot evaluate the deep neural networks Leela uses, as GPU does.
People actually waste their time working on computer chess? Jeez. Get a life.
@@jx14aby what an awful thing to say. the founder of AlphaZero recently won the noble prize because the work he did developed the nueral nets for AlphaZero were able to be applied to predict protein folding with AlphaFold. It saves researchers thousands of manhours and allows for massive advancements in the medical industry. If he didn't "waste his time working on computer chess" the world would be deprived of of the greatest achievements in recent history.
@@jx14aby your ignorance is off the board
Well, Leela on CPU can still beat grandmasters easily. Just will lose a few hundred ELO from having the nodes searchable being reduced by about 2 orders of magnitude.
That's what Agadmator said, no?
Stockfish: I got this (easy checkmate in 1)
Leela: you sure ? 👀
I feel like these two are playing nothing but engine moves.
Interesting...
Let's start procedure
Not many engine like moves in this game surprisingly. Maybe the black Q to a1, maybe... A lot of really bad choices in this game, any chess player can see right away. More like a composed game really to come to this draw. Strange game. Engine games usually are strange, but in an incomprehensible way. There were blunders in this game too, which is unlikely for engines. Ok, there's blunders in every game you can argue, even in eng games, but when human chess players can see them... I call fraud in this game.
If I were to play those moves I would be checked for beads. No standing ovation.
Leela : Super Tal
Stockfish : Super Magnus
Alpha Zero : Super Morphy 🗿
Stock"Fish" has to be Super"Fish"er.
Stockfish aka Magnus is better than all
That's why it disappeared 😂
isnt leela the same as alphazeri
@@dariogreggio7981
Leela is the home brew version of Google Alpha Zero. At no point has it had the computational brute force available to it, to match Alpha Zero.
Incredible clever last set of moves to force stall mate. Thanks for sharing!!
This game shocked me. Leela was so forward thinking that it saw winning was not possible and orchestrated this outcome a dozen moves in advance and stockfish seemingly was helpless and unaware due to its own single-minded ruthlessness to be aware of Leelas strategy.
10:32 even leela is sacrificing the rook 😂😂, levy would feel so proud
Stockfish : I’m Stockfish
Leela: You’re Stupidfish
Leela: You're Stallfish
Stockfish: I have you trapped
Leela: Hold my rook
😂😂😂
These algorithms seem to have personalities. Anybody know why people associate Leela as a she/her but stockfish has no associated gender? Is neural transgender a thing?
Beczuse Leela has a female name and stockfish is a fish 🐟
I once gave rook odds to my friend, but he played well, got a passer and when position became unwinnable I sacked my other rook and the queen for the stalemate. He was looking at the board for another 20 minutes. It could be his first win in years, but I wiggled out. Still the best game of my life.
Leela went like 'I would give up my King as well if I could" 😂
These engine games are the best, more of this please! Makes me feel so inferior!
Antonio’s Brother in law also has a Chess channel Jozarov Chess he shows many Stockfish and engine games as well as the top GMs and does a great job as well
@@michaelmassaro4375 thanks, will take a look!
@@michaelmassaro4375Yes, Jozarov is the best (:) at commentating engine games. He sometimes goes through like 20 different variations/sidelines but somehow never loses me throughout. Great to learn from his presentation of the ideas in the game
@@michaelmassaro4375 wait, Jozarov is Agads brother? Good to know 😅
@@LeonardoContreras-r1k Brother in-law (Agad's wife's brother)
Anish Giri's Dream. Brilliant Tal-like sacrifices to make a draw.
Pure poetry! It taught us that if we're losing and we can sacrifice everything getting a stalemate, this is one of the most "simple" ways to get a draw💪
it's interesting to check both yours and levy's analysis of this game
Stockfish: I'm gonna checkmate you..
Leela: Nope, I'm gonna stalemate myself...
Stockfish seems to remember what Tal said once when asked why he gave away a Pawn---"It was in the way!!""
11:08 Technically, the move ...Qf2+ ends the game immediately, because it produces a "dead position" as defined by the FIDE laws of chess, 5.2.2.
Good for FIDE games
Pretty sure FIDE doesnt govern computer chess
So I am a SUPER new chess player, and it will probably sound very naive, but is the game a draw because the black pawns cannot move and ANY move the king makes is checkmate? I didn't even know this was a rule, if I was playing my uncle I would assume this was checkmate because the king would HAVE to move into a death position... I guess that's not the case... Fascinating!
Yes. It’s called stalemate.
No. It's because black has no legal moves
Yeah this is one of the more unfair seeming so-close-yet-so-far type situations in chess. If the king is not currently under check, and cannot move to any square where it isn't at check, while also having no other possible legal moves, it is called a stalemate, and considered a draw.
It's actually one of the things top players are trained to be wary of in end games where they have a massive advantage.
@@arjun_ragafanatic I see, pretty fascinating...
Honestly this is one of the greatest games ive ever seen😢. We see miracle wins all the time but you almost never see crazy stalemate draws. Unbelievable 😍
The narrative on your videos gives such great context, I miss it sometimes. So informative
Agadmator should get a custom engine made named, "uncle from the bar & library" which of course only plays uncle moves.
Leela remembered us a lesson often forgotten: in every position, the key is the king. You should always start evaluating a position after understanding if your king is good or not. After all, the king is a tactical weakness. In this game, white sacrificed for the whole game the safety of his king in order to attack, while Leela tried to play solidly without giving too much. Stock fish only had a mating treat in the last move he played, and that’s exactly where Leela showed us the importance of checking your king position.
“Mr stockfish they found a phone in the restroom, we believe it’s Lela’s.”
Stockfish : There’s no doubt, begin the procedure
A Rosen trap of epic proportions!!!
Eventhough its not impossible but when i saw the title i thought Stockfish lost with white pieces and i really amazed. But anyway draw is acceptable. They have huge depth of calculation and seeing these lines are possible. Thanks for this great game Agad.😊❤
Stockfish loses with white depending on the opening.
@ndnd7614 yeah I meant openings that at least doesn't have disadvantage.
@@hosseinmohammadi8776 I remb I saw a game where stockfish lost as white when the opening evaluation from stockfish was 0.0. So it not necessarily disadvantage position
@ndnd7614 wow, interesting.
Eventhough a bit unbelievable.😊
Maybe a blitz game or I don't know anything can happen but forexample stockfish 15 with that strong depth Plus nnue , for me it's strange. But I have to know about it more. Thanks for your statement. It made me to go and search more about it.😊🤚
There was also an amazing game from another RUclipsr called ASMR Chess where he bought a super old chess machine called Mephisto (like from the 1980s old) and put it against Stockfish. The wise old Mephisto also held Stockfish to a draw! Granted, they started from a normal, fresh game instead of pre-programming to start in a Caro-Khan, but it's still pretty impressive for such an old computer.
What a video! TLDR, As a data scientist, what happened here (for people that doesnt know too much math), suppose that you are going from a city A to B with a car and think thats the fastest way to do it is take C1 runway (100 km), but sometimes it gets lots of traffic using time to travel about 200 km, and C2 (150 km) doenst get traffic. We can decide into C2, but stockfish actually doesnt. It wasnt a lc0 art, since stockfish knew the drawish lines. But this didnt make this video worse at all. If my hypothesis is correct, this game shows perfectly the psychological factor in chess. From the move 38 Nxb5, the game is completely drawn but many people couldnt find the line after being positionally crushed for about 40 moves
6:55 I dont know how lc0 evaluates (if engines also generates a log from those matches...) but this can be a "bluff". Lets say do you have 5 suboptimal moves, maybe do the "best" supoptimal leads to a easily eval (lets say that best suboptimal leads to material loss in 5 moves and others in 10), so its better to do something thats get you worst, but it takes more moves (and more chance of errors). 6:40 is the same concept, how 3x0 is better than 3x1? it isnt. the difference is, stockfish found the trick on 6:40.
I cannot prove it, but can give some decent imagination. After 28. Qc5 stockfish (depth 33) evaluates 5 moves (ordered) for leela (b6, Rd5, h6, Rfe8, Qc8), scoring from 1.37 to 2.19, with the next move for stockfish (ordered) (c6 a6 O-O-O O-O-O O-O-O) When leela executes 28... b6, the 5 lines are (c6 1.42 cxb6 0.11 Rg3 -1.00 Kf2 -1.16 a6 -1.21). If 28... Rd5 was played, all lines are +1 for white. So 28...b6 and 28...Rd7 gives the same evaluation locally, but b6 allows some "chance" of blundering.
Finally, lc0 pratically loses by brute force, but the little piece of "human playing" saved lc0, since it doesnt not eval how to win, but how to not lose. This could be a good test for stockfish developers: how to make the engine, when completely winning, avoid chances of not winning by this kind of situation by, as an example, giving some material.
Again, cant prove it, but as an example: Even at lower depth (33) the drawish lines starts to appear after 38 Nxb5 Qb6, another line is 39 Nd6 Bxd6 40 Rxd6 Rxd6 41 Bxd6 Bxd6 42 Bc5 Qb5 53 Kf2 Ba8. Completely drawn (all evaluations for stockfish in this line are play or get worse and the result is always +0). Instead. 38 c7 didnt raised any drawn line (depth 33), always giving +2 for white.
This problem can be saw as finding a local optimum that wasnt the global optimum. Without deeper analysis (this took something like 15 minutes for me) stockfish blundered with 38 Nxb5 (even being the best move), leading a superior position into a drawn state. Thats why some algorithms like simulated annealing, genetic optimization etc are essential.
I really enjoy watching these strong engines play! I can just enjoy the great chess without having to care who wins. And extravagantly good chess this was!
Me: *finds the move that a 3664 engine played*
Also me: *falls for a trap by a 1200 player*
😂 this is great
02:00 which specific “exact same opening” would they “constantly play” if not compelled to do otherwise?
Reti
Yeah, I don't think chess engines work that way. The same engine does not always make the same first moves. What does Agad mean here? 🧐
@@Think_1234 Actually they do. They play 3 openings max.
King's gambit
@@kingStribor 3 is not 1. And 3^2 is not 1. So they are not playing the exact same opening. Instead they variate between different openings. Even if between only a few.
it is really amazing how much poison there is in the position at move 41-42 even with such a dominant position a check mate is possible with a simple mistake !!
Thank you. Chess is amazing, especially on the highest level.
Reminds me of a tactics trainer puzzle!
After this game a Stockfish engine upgrade came out that foresees three-piece sacrifices leading to stalemates lol.
I love engine chess! Only thing better is human versus engine
Really wonder when precisely Leela and Stockfish both understood what what going to happen.
Like… Leela must have figured it out way before Stockfish
Seemed Leela accepted that it (she?) can’t win much earlier than Stockfish thought Leela would
@@507RiverRockyeah but I wonder, could Stockfish have spotted this earlier and leveraged it’s dominant position and when did Stockfish started computing those absolute GOAT moves
Not necessarily. Stockfish could not have understood what was going to happen when it played Nxb5, because it had alternative moves which preserved its advantage. But Leela might not have understood what was going on at that point either; it might have just kept playing the only move that didn't lose immediately, without knowing that it could draw.
@@JohnDoe-ti2npyeah that’s possible I really wonder that’s why I would be really curious if there is anywhere we could see the rating each engine give at the position on the moves to see when each see that
Stockfish must have overlooked Rxc6 or Bg7 not calculating far enough
So, even the immortal Stockguppy falls for stalemate tricks. Wow 😮
Stockfish should’ve used its Spider 🕷️ Sense hey Leela giving me pawns for free ?? What up with that but nope super nice moves from Leela disposing of its property
This is truly brilliant what Leela has played against Stockfish, sacrificing pieces to draw the game by stalemate. 👍😇
Excellent content!
Superb chess!
An amazing stalemate! Thank you for sharing! How deep is that?
Hey Agad I love all your videos on going over chess engine games, very educational. I was wondering if you've heard of an engine called Patricia, I feel that could make for some good content and games to go over
The title should have been( A bolt from the blue).
The question is after which move did it plan this, so srockfish didnt notice it.
I think it's the first time I see a game ended in a draw in this channel... I very much liked it and I think it would be great to show more games ended like this, sure there are a lot that have been interesting between humans too (greetings from Ecuador)
@9:15, I wonder what was wrong with queen takes knight+check ... I guess it would still be stalemate but less of a guarantee of one.
Thats something i want to see more of - beautiful brilliancies from games that saved a draw when it looked like a game is giaranteed to be a loss
Wow...Whatta goosebump moment....Leela the tricked Stockfish 😂
Amazing! Great game!👏👏👏
Thanks again for another informative episode.
ok ... it's been a long time a game has left me this flabbergasted. holy
The chess computer on my phone does this to me all the time. By the time I realize its chasing me down with a piece I simply can't capture, it's already too late. For humans we don't intuitively look for this, but for a silicon brain its just yet another permutation that's part of its calculation methodology.
uff, Ive never even imagined this kind of forced stalement, absolutely ming boggling
Look @8:02, the positions reached when machines are playing are out of this world (literally).
Thanks for another engine video. Please do smth like that more often.
Amazing!!! Thanks!!!
what is the line that top engines take if you leave them to it?
@agatmator at 3:36 couldn't white play bishop f4 and traps black's queen?
The way these two play sure is... interesting 🧐🧐🧐
At 8:47 why not take the white rook at D2 with the black rook?
Wow just wow 😮😮😮 this level of chess is incredible
Is the botvinnik carls variation of the caro named after Carlsen? I didn’t know that or did Agad just say that around 2:30
I've sacced all my pieces to make forced draws a few times (though not 4 of them as here?), but I didn't celebrate it, because I was usually uncertain as to whether doing so was best.
This reminds me of my first time playing chess. We were in High School and I just learned how to play chess. I am always losing to my friends but I accidentally discovered stalement. They were amused by that time because they did not expected a newbie to pull off a stalemate. Since that day I think they fear me for playing sacrifices because I have been known for stalemates. Good old times😅
What is the score between stockfish and alphazero?
Would you mind checking out Bagirov vs Gufeld 1973 the “Mona Lisa” game.
Unreal game!
I'm fairly new to chess here. So am wondering: after Leela played Pb5.....why didn't Stockfish take Pb5 with the knight?
That might be the craziest draw I’ve ever seen
I know I am missing something but can someone explain in 8:48, isn't that a free rook for black in RxD2?
it is a free rook, but if R goes to D2 then Bishop goes back to D4, then if R took takes d4, then Queen to f7 check,
Outrageously brilliant.
Can anyone explain why in this position 4:35 bd7xa4 is not played by Leela?
Hi. It may be because Bd7xa4 sets up a self-pin (the bishop is now pinned against their own Queen on a7 and White has a rook on a1). If the Queen moves, the bishop is now en-prise, so Black effectively has two pieces out of action until the pin can be resolved.
@@TJConcept1995But ra8 can be played in that position? Am I missing something?
@@Timepasser_me Yes, ..Ra8 can be played, but the rook is behind the Queen, so it is still pinned.
If things were the other way round, i.e. Queen on a8 and Black could play ..Ra7, then that would work as it would break the pin and still defend the bishop on a4. If the bishop on a4 was then moved, Rxa7 is met by Qxa7.
I was considering adding the Caro-Kann to my repertoire but if this is what it takes to earn a draw, I'll stick with some other openings.
Chess is Amazing! Nice find!
Leela chess zero is the alternative of alpha zero made by the same programmers so that just means if stockfish is better than leela its also better than alpha zero
Beautiful. Basically a win for Leela for coming with this brilliant strategy and Stockfish falling for it
You haven't mentioned which move is the strongest move recommended by the engine.
at 8:49, why lela didnt capture hanging rock?
nvm, its mate in 1
Wow. What a game! ✅✅✅
Wow, that was amazing 👏
3 agadmator videos in one day??? We eating for real
what is that opening the machines would play over and over?
Hi agadmator!
Quick question about Chess engines: Can engines know if the game is a chill training one vs a tournament one?
I would assume a good strategy might be to practice uncommon/weird moves that the other Algo might not have checked. Assuming algos didn't to full breath searches and they trim certain options, playing such moves might being the game to a place where the other Algo might be at a disadvantage.
Thoughts?
The beauty of Neural Networks are they learn from their mistakes.
6:15 What? Did you say the “NNUE principle”? What’s that
How long does it take for the engines to play a game?
Wow just giving up pieces to draw via stalemate,crazy 😅 🤣 if some player pulls this off in a live classical game it would be incredible 😊❤
Incredible!
What is the "NNUE Principle" ?