Leela Plays An Immortal Game Against a 3300 Engine
HTML-код
- Опубликовано: 25 фев 2024
- Download Mproov and Improve Your Chess Today! app.mproov.me/AgadRUclips1
Search all the games on my channel here! agadmator-library.github.io/
LCZero 0.31-dag-5c1051f-BT4-6077500 (3636) - Tucano 11.14 (3293)
TCEC Cup 13 Round32 (tcec-chess.com) [1.46] E21 2024.02.26
E21 Nimzo-Indian, three knights variation
1.d4 Nf6 2.c4 e6 3.Nc3 Bb4 4.Nf3 Ne4 5.Qc2 f5 6.g3 c5 7.Bg2 Qa5 8.d5 Nxc3 9.O-O Ne4 10.a3 Qa6 11.Ng5 Nxg5 12.Bxg5 Ba5 13.b4 Bc7 14.b5 Qb6 15.a4 a5 16.e4 h6 17.Bd2 O-O 18.f4 exd5 19.exf5 d4 20.Bd5+ Kh8 21.Rae1 Qf6 22.g4 d6 23.Kh1 Bd8 24.Be6 Bxe6 25.fxe6 Be7 26.f5 Qh4 27.Re4 Bg5 28.Be1 Qh3 29.Qe2 Qb3 30.Rf3 Qb1 31.Kg2 Be7 32.Rf1 Qb3 33.Ref4 d3 34.Qe4 Nc6 35.bxc6 bxc6 36.R4f3 Qa2+ 37.R1f2 Qxa4 38.Qxd3 d5 39.Rc2 Rab8 40.Kh3 Rb7 41.Bh4 Bf6 42.Bxf6 gxf6 43.Rg3 d4 44.Re2 Qa1 45.e7 Re8 46.Qd2 Kg7 47.Kh4 Qb1 48.g5 hxg5+ 49.Rxg5+ Kf7 50.Rh5 Rg8 51.Re1 Qxf5 52.Rxf5 Rh8+ 53.Kg4 Rg8+ 54.Kf4 Rxe7 55.Rh5 Rg7 56.Rh8 a4 57.Qa5 Re5 58.Qc7+ Kg6 59.Rg1+ Rg5 60.Qxc6 Re7 61.Rxg5+ Kf7 62.Rh7+ Kf8 63.Qxf6+ Rf7 64.Qxf7#
Timestamps:
0:00 Hello Everyone
Check out agadmator's merch here teespring.com/stores/agadmato...
teespring.com/design-with-a-s...
----------------------------------------------------------------------------------------------------------------------------------------
If you realllly enjoy my content, you're welcome to support me and my channel with a small donation via PayPal, Bitcoin, Litecoin or Nano.
Link to PayPal donation www.paypal.me/agadmator
Bitcoin address 12VEbMQPyLzBoZzw9yuNofph4C9Ansc4iZ
Litecoin address LbSuZuBffDCNmr5CSZbY7W2zM83w4ZvnC7
Nano address xrb_383y7ofu5wsyfr9o8rh93aqaq8aixpdcbaud5iubydukz5moiadsirmuzgoq
Check out some of the books I enjoy www.amazon.com/ideas/amzn1.ac...
Check out ALL my videos here • You Snooze, You Lose |...
Steemit: steemit.com/@agadmator/
Facebook: / agadmatoryoutube
Twitter: / agadmator
Instagram: / agadmator
Lichess: lichess.org/@/agadmator
Chess.com: agadmator
League of Legends: agadmator (EUNE, my friend is using my EUWE account for a couple of years now)
Blizzard: agadmator #2992
Check out my Hearthstone channel here / @agadmatorsotherchannel
"Watch me without ads on your Amazon devices (bit.ly/Agadmator_Amazon) and Roku TV (bit.ly/Agadmator_Roku) 
Развлечения
Leela doesn't use NNUE, used only by the classical engines, which use alpha-beta pruning, to advance their evaluation function using efficiently computed on CPU neural networks. Leela uses Monte-Carlo Tree Search with the deep neural network, which uses a GPU to perform an inference
Went over my head
Smjh me ni aya kuch but accha lga
goo goo gaa gaa
Wow didn't know lc0 uses monte carlo search
@@skorpid-tm4531 It's based on the AlphaZero papers
"There's always a bigger stockfish" - Qui-Gon Jinn
Not really Zero is too deep for stockfish sometimes. This game remind me of a game were stockfish won against Zero with 120 movies, while Zero took only 50 movies to win with the same positions.
Reminds me of the least turns competition for cubing lol, forgot how they did that at google but somehow they proved it only takes like 20-23 moves to solve most Rubik’s cubes or something lol. Fascinating how much of a future these games still have even with computers who can spank us to pieces. It’s still just as fun
You are stockfish and this joke was by Zero @@JohnJohnson-fn2we
@@benr3799they found a Scramble which could have been only solved in 20 moves and then found for every other possible Scramble a solution of shorther or equal moves
@chickenwheel45 I came here to comment exactly that! 😂
After tucano took the free piece, leela said " yepp that checkmate in 79 moves" 😂😂😂😂
Yep, probably engines will either always draw or always win with white in not far future.
@@woody442 Always the chance that when/if chess is solved that black wins, who knows
@@905LilO Quite possibly. Or could come to opening move alone. For example 1 E4 always wins whereas 1 E3 is a draw.
@@905LilOchess will never be solved, because the moves possibility are too much for even an engine to analyze completely
@@905LilOchess is unsolvable, there are more possible moves than molecules in the universe
Paralysis seems to be a common theme in engine games.
full blockade
only way to win cuz the other engine sees all the "obvious" stuff xd
For AI engines, yes definitely. They utilize the untaught fact that a useless piece is indeed worthless. While a human derived engine will panicky hold on to material regardless of its dynamic value
Yeah I've seen Leela put opposing pieces in lockdown multiple times now and I imagine as the engines get more advanced that will be the constant theme. Activity is the most important thing in chess in my eyes for this reason as undeveloped pieces that can't interact with enemy pieces are worthless.
@@ninja8flash742 They are also the interesting decisive engine games compared to the ones where it is not remotely clear why one side should be favored until a 100 move long endgame.
"I like it when I crush another engines ego." -probably Leela
Bobby Chess Zero
Me: What is best in life?
Leela: To crush Stockfish, see it melt down, and to hear the lamentations of its programmers.
@davidanderson2357
[Laughter] "Yes! That is best in life."
😂😂😂 nice one
No. Definitely Stockfish. He runs rings around leela most of the time.
Please show more of these gigachad engine games. It's like watching chess from 200 years into the future or from a different galaxy. Utterly fascinating.
"1858 and 3058, at the same time" if that makes sense.
Looks more like Morphy chess to me.
I wouldn't watch engine games even if someone paid me for it. Most boring stuff ever 🙄
Honestly, it's not even the same game as human chess.
So how did you even get here then??@@fundhund62
that however he sounds rambling at times with this meandering discourse on the game it is intriguing however be careful :D
"Not just with a grain of salt, but a huge plate of salt." This is now my email signature.
But that much salt is not good for health😆
That advice doesn't make sense. If something calculates better than you, you don't have any salt to throw, the only smart thing to do is accept it. But since a better engine might exist, you should take it with a grain of salt until you are sure you have the best engine. If it makes a rare mistake, not like a human would be smart enough to correct it
rofl :D he knows what he is talking about however calls the AI a person someone playing chess hahha
Tucano: haha I won a piece
Leela: haha you can't use any of your pieces
WHAT IS WITH THIS website it bugged and did not back button correctly TO THE VIDEOS and replayed HELLO EVERYONE 3x rofl THAT IS FREAKING CRAZY :D rofl :D
He shouldve just sack a piece instead of trying to save everything.
Im sure Leela calculated that too though..
The most insane part about watching these supreme engines is that everything they play is so aesthetically pleasing. In the past, engine moves look sloppy but they worked anyways. Everything Leela does looks beautiful.
If you like drunken Axl rose
IM (2400) -> GM (2500) -> Super GM (2700) -> Overlord Engine (3300) -> Supreme Overlord Engine(3600)
It feels like DBZ. Always a new boss overlord that raises the ante comes up making the previous one feel like child's play.
There is no super GM
You forgot Hans Niemann (9999)
This sounds like something I would have written when I was 12 and cringe over today.
M??
My dream is to sacrifice a piece willy-nilly and a few moves later half of my opponent's pieces are completely useless. That would be incredibly satisfying.
What's your opinion of Armageddon w/ auction, even in classical and especially in 9LX?
---
Part0of3 - Intro in CSQPod:
A - 2:22 in B2Gi-6EgcfY ( clip for csqpod & MVL ) and
B - rc-GquDy3PU ( CSQPod & hikaru )
To break up ties in classical 1v1 matches or knockouts ( as opposed to breaking up ties in round robins or swiss )
- MVL & Fabi don't want super long armageddon's to break up ties
- Yet Hikaru & Fabi don't want rapid/blitz to break ties.
Compromise: If you're going to do 25+10i rapid tiebreaks, then why not 25+35i classical armageddon tiebreaks?
---
Part1of3 - If the point of classical 9LX is that, as said by the top player in
- classical-time classical-variant chess Magnus, Carlsen, 9LX isn't suited to rapid/blitz
- classical 9LX Wesley So, players don't play so consistently in rapid/blitz 9LX
Then classical 9LX should have classical tiebreaks.
Generally I'm known to have a huge bias against Magnus Carlsen the chess player (for being a hypocritical cheater), but Magnus Carlsen the 9LX player isn't so evil: Magnus is setting up classical as in 120min+ WITHOUT consecutively repeating starting positions (as FIDE had hypocritically insisted on doing in 2019 & 2022 WFRCC).
---
Part2of3 - An armageddon with 120min+ for white might be too long, but there IS precedence of a 60min+ armageddon won by Gata Kamsky in an old USCC in 2010 and even sorta in 2013, I think.
You can also do like 25+35i or even 10+50i instead of 50+10i or 60+0i, all of w/c all equivalent under FIDE's conversion. Rapid games are sometimes 25+10i, so just increase the increment by 25 seconds.
---
Part3of3 - And I believe Bobby Fischer would like armageddon too (at least in 9LX if not chess).
Bobby was preaching about pre-arrangement, but I don't think increments & 9LX are enough to solve this.
In xiangqi, their armageddon has auctions and was invented by a Ronghua Hu known as - surprise surprise - the Bobby Fischer of Chinese chess!
- pKdQYb
Armageddon w/ auction makes every game fair & decisive.
I believe this'd mean
- A - no more draws but in a good way not in a no stalemate way. And a fortiori no pre-arranged draws
- B - no more rapid tiebreaks. You can play chess w/o Armageddon and then settle classical matches w/ classical tiebreaks
- C - Can have best-of-1 knockouts
- D - All single round robins & single swiss are fair.
- E - There's no more issue of drawing a best-of-4 by compensating for a white loss with a white win & 2 black draws like in 2016 & 2023 classical wcc's where Sergey & Nepo had more black wins than Magnus & Liren.
Plus, at least in rapid, bids for black in 9LX go like 12min/15min while in chess they're like 9min/15min since 9LX doesn't have as many draws or as much white advantage. I expect classical armageddon bids will also be lower in chess than in 9LX.
I notice these 3 improvements of Bobby
- 9LX
- Increment
- Armageddon (w/ auction)
Are used more in rapid than classical but they should be used more in classical!
They also eliminate the top 3 worst things about classical-time classical-variant chess:
1 - theory
2 - dirty flag
3 - draws
@@nicbentulanwtf
Tal checks his watch. Time to sacrifice another piece!
Imagine the commentators of such a game: 'They play THAT move?!!! What are they thinking? That surely must be losing. So what does the Eval bar say?...... O.O'
@@nicbentulan dawg this is an engine game
Leela is playing an endgame during the middlegame. Kh3 Kh4 with the idea of Kh5 to capture on h6. This is beyond comprehension!🤯
Great observation!
Some of the Alpha Zero games against Stockfish show it can achieve a zugzwang with Queens still on the board : something considered impossible for humans.
(It is recommended the Queens are exchanged prior to going for a zugzwang in human play. )
@@davidanderson_surrey_bc Yeah, just pushing the king up into that pocket to do some work like any other piece, while the opponents queen and rooks are still on the board.
Even preparing the possible trade of bishops on h4 with Kh3 is incredible before Kh4 is even played. The confidence is incredible.
@@michaeledwards2251stockfish > leela
It's an astonishing game. Sacrificing a piece to gain a kind of unstoppable momentum is a pure, computerish idea.
There will never be a human who could play like this
Sacking a knight to paralyze both bishops, a rook , and the queen is the kind of trade in the spirit of the old masters but executed in the future. simply beautiful
I doubt that. Someone will eventually be able to play 3300 . Knowing the other side best moves will eventually happen from a human. But it will take about 30 years for someone to spend their entire life to chess to achieve. Beating Leela, I agree. But a human cam memorize alot. There's no limitation on the memorization ability of a human so meeting that can happen. Won't ever surpass. It's the challenge the humans will rise to. Time and time again it's happened
@@kenw2225 Leela will look like a club player in 5 years.
@@kenw2225 10^180 possible chess positions, a massive amount more than atoms in the universe by a factor of 10^100 (ten followed by 100 zeroes). Totally impossible to memorize, absolutely ridiculous suggestion from you. Perhaps humans will eventually surpass 3000 elo, but it would never be through brute force.
I really love how this Leela plays. Suffocating your opponent, in stead of fighting him...
Putting opponent's pieces in jail - signature Leela style 👽
HJahaha well observed my friend.
I love so much these videos of engines playing. I would love to see you showing more and more of them! Thanks, agad.
It feels like DBZ. Always a new boss overlord that raises the ante comes up making the previous one feel like child's play.
Or like Cell evolving as he defeats opponents!
Power level 9000!!!!!
"This is only the beginning of black's suffering." I for one, am glad engines can suffer. I did not know this. But it brighten's my day.
The white march in this game is one of the most terrifying things I've ever seen
That was one of the most mind blowing games of Chess I've seen in years.
we want more of Leela and IA neural networks games ... they are exciting and include many new ideas unexplored by humans
Would be really interesting to see the evaluation curves from each engine throughout the game: at what point did Tucano realise it had lost? By how much did Leela think it was winning at the point if sacrificed the knight?
Note that while Lc0 uses neural networks, it uses a deep neural network rather than NNUE that Stockfish and Tucano use
The two knight sacrifices in this game reminded me of Averbakh, Yuri vs. Spassky, Boris USSR Championship Leningrad 1956 Round: 3 ECO: E74 when Spassky played 16. ... Nc6!! to get a tiny bit of counterplay after botching the opening.
Those were the days...
Best RUclips chess Channel on the planet
By a fucking country mile
No 24/7 Magnus Carlsen hype LOL
Gothamchess is for kids
Leela's full name is beautiful!!
She's hot af
I missed these engine battles. I've beem waiting
More engine game please :) This is pure perfection and poetry :)
Oh how I have missed these engine vs engine games
Reminds of AlpaZero with the blocking of the undeveloped pieces
That was incredibly beautiful and terrifying for humans to comprehend all at the same time. Absolutely mind-blowing.
this was so entertaining! more engine games please!
Great video. Thank you
Leela sacrificed a piece in the opening for a big lead in development and to gain space. Ng5 traded black’s only active piece. At 6:23 black struggles to untangle and develop any queenside piece. The queenside pieces are also bad pieces because other pieces and pawns are blocking their activity. At 10:17 black gave the piece back since the queen being the only active piece is not enough. The bishop is tied down blockading the e6 pawn so Leela is still better.
I've been following Leela since she was young and only a few blocks tall (😂) and I still am amazed by her games and will never get tired of them.
Brings to mind Alpha Zero's Immortal Zugzwang game.
I love your videos for 5 years now Agad!
Its good to see these engines play another every now and then to see what the grandmaster train against.)
That piece sacrifice in the opening was something else.
I really enjoy the logic of programs.
The scariest part about engines is how they operate logically beyond even numbers calculation. It sacrifices a peace to render multiple pieces of the opponents useless and effective unplayable. Makes you think if AI ever went rogue it would render humanity helpless before you have seen it coming just like tucano engine was.
Not all engines are "ai", for example stockfish is just regular algorithm. Furthermore something like "ai" doesn't really exist. Those engines are just neural networks which are in essence fed by large amounts of data. They don't make any logical connections, nor play chess like humans, they just choose the move that makes the most sense based on data they had before. 2000 years ago people like you were scared of eclipse, because they didn't understand it, now people like you are scared of ai for the same reason.
I like when you do extremely precise chess engine videos. Brings me back to the Alpha zero series which were my favorite.
This reminds me of when Leela first burst onto the scene and destroyed Stockfish. When you look into the games highlighted in the Gamechanger book, a constant theme was that Leela would be down material but would practically immobilize Stockfish. Having 3-4 pieces bunged up, and typically on the queenside, was an absolute hallmark.
Sometimes I wonder why Tucan didn't sac the knight sooner (for example after e4, play Nc6. It forces the exchange and restores parity, plus gives you some room to breathe. Leela has sacked a whole piece for the position. It's obviously objectively wrong (Tucan is 1K points higher than me!!), but for an intermediate human that feels like the right approach to give back the material, what else is better?
Wow! Insane! Thanks!
Awesome to watch.
Hey @Agadmator. you mentioned such a good point on 5:50 . A really nice point about the new AI Models, is that they are trying to Let the computer build the knowledge about chess themselves. they instead of letting the machine crunch the depth the choose another method.
A new research paper has found a really beautiful idea, demonstrating for the first time that it is possible to train a neural network to play chess at a grandmaster level without relying on explicit search techniques. This finding challenges the long-held belief that sophisticated search algorithms are indispensable for mastering complex games like chess.
so imagine somedays with Quantum computers there will be a new Era of chess. maybe transforming the game completly.
Always amazing to see Chess at this level.
absolutely incredible
“Pawn to b4… as it often is the strongest move…”. 😆👏👏
At 1:49 finally learned how to correctly pronounce Agadmator! Antonio spoke his nickname!
I've been in the Leela discord for a while now, and I'm pretty excited for BT4, as initial testing shows that BT4 and BT3 have beaten the development version of Stockfish. Hopefully, we will get back the TCEC crown once more!
Where to get the discord link
Not ture. SF is still better than Leela in TCEC hardwares ratio. Only if she improved +20 +30 in 2 months , Leela will win.
what is BT
@@nabildanial00 I believe it standards for batch transformer, but I could be wrong. It's a transformer-based neural network, and it's what GPT3 uses for it's large language model.
So, you can take the idea of how GPT3 generates text based on the context of the content that a user writes and you can apply it to BT3 and BT4. Only, in this case, it generates a move.
@@drmrboss testing has shown that BT4 is 35 Elo points higher than the latest stockfish development version, so it looks promising.
100 games is a lot and thy definitely did not play that many. Leela won 3-1 and advanced to the next round of the cup.
This was part the 1st round of cup13. Only 2 game pairs (2 openings, engines get to play both sids of each opening once) was played by default unless a decisive lead is reached before that (at least 2.5-0.5). If tied after 2 pairs, extra pairs will be played until a decisive pair is found. As time control is 30m + 3s, there's not really resources to be playing 100 game matches in a elimination tournament with 32 participating engines. 100 game matches are reserved to the most important events like the superfinal as it would take 4 days with this moderate time control.
that castle by white after Bxc3 is beyond my wildest dreams
Thank you for your take on engines. I find it strange that so many people believe that "stockfish says" is not only objective fact but actual knowledge that they themselves posess.
The description of innacuracies, mistakes and blunders are taken as some kind of gospel.
Some so called inaccuracies put your openent in a position where they have to play five or more precise and difficult to find moves, that is most often far more likely to end in a win than the suggested move that can put you in that position.
They are no doubt very usefull tools but the plate of salt is a great description.
Yeah, the engines evaluate the position as though they're playing against another engine that never makes a mistake. So if it's in a losing position and it can either A) Make a move that simplifies to an endgame that your uncle could win but postpones checkmate for 20 moves, or B) Make a move that will result in getting checkmated in 7 moves but only if the opponent makes 7 extremely precise moves and otherwise is actually a draw, it'll play A every time.
But is hoping for your opponent to make a mistake good chess?
@@thebcwonder4850 That's what chess is. From the start of the game, it's a draw until someone makes a mistake. Your goal is to complicate things enough that they make a mistake and you can win.
@@thebcwonder4850 Putting them in a position where they are likely to make a mistake is 50% of the game, the other half is not making any yourself. I would say that if you win, then you didn't make any "mistakes"
I love a few engine games a week. Would like to see more!
I love these Leela games ☺
I love when games speak to you.. This tells me that even with the best tools we've ever built, we have not even scratched the surface of what can be known, even outside of the world of chess. The "correct" decision will never make sense from our limited radius of human knowledge.
The most efficient way of generating "brilliant games" is by using two engines with a 300-point rating difference. The weaker engine simply does not see the traps of the stronger engine, and yet, from a human POV, it is not committing any blunder
I started watching chess contents with Alphazero games from agadmator in 2018
the crazy thing tucano held for 79 moves.
Please always keep that picture of Cortana for Leela. It provides such a great atmosphere of mystery and intrigue!
You did a good job with variations. Thank you!
What do you think of nDWpjxXYLZ4 ?
At 3:47
My uncle 👁️👄👁️
i dont know if i would call it an IMMORTAL, as there wasn't a full-out material sacrifice that rendered white with a few pieces used to checkmate the black king - but it was one of the most amazing games i've seen in years. Thank you!
Nice, the engines and Morphy are my favorite games.
Love this image for Leela. What a game.
It sees two wildly counter intuitive moves demolish the show ... then it rolls the plan out like a glacier..... then it just plays 'H1'. Beautiful. Not human, but beautiful.
please def. do make such a video comparing the strategies employed in different models
Thank You for mixing it up
Nice quote at the top by Lasker
That is a super dedicated father-in-law!
Wow. I was laughing the whole video. Those moves are insane. And to think that they were played against an engine that would obliterate any human is mind blowing.
Thanks for sharing.
Wonderful game!
Please show more games from Alpha Zero or Leila or other engines.
Those are awesome 👌 ❤
Thanks
I've actually for some time hoped to find a good video about how the cutting edge chess engines actually work and how they differ. The absolute best thing would be to hear about that from you!
Thank you,Agat....
Please, continue these series of engines.Any
How to say? Leela plays chess kind of like a Go master . It didn't matter that Leela went down a piece early in the action because 3 of black's pieces remained undeveloped and out of the game for way too long and even black's queen was boxed in and isolated for a while. Such superior control of space on the board is so Go like.
Leela is based on AlphaZero which used the technique first used by AlphaGo in playing Go
Always great when we see 3600+ chess from Agad!
Cheers everyone! :)
Agad shittalking my uncle again 😂
13:51 And it was in this position that Tucano did not resign, rather it allowed Leela to completely demolish it with a checkmate...
interesting at 9:34 if black takes on a4 instead of pinning the bishop there is going to be a slaughter starting with h4 i guess
Mathhew sadler has a great RUclips channel called silicon roas focussed on engine chess
I notice that the best engine games often focus on position and gaining SPACE in a way that humans do not.
Even this guy had no real idea of what he was going through
Really enjoyed. I was laughing throughout the game. Wowww!😅
That's signature Zero-engine play. Sacking a pice to lock the rest away.
16:25 We all knew it wasnt because of beauty, but because of winning b4 move LOL
I love that throughout the years you've kept making Evans jokes when someone (or something!) plays b4 lol
I would love a playlist of immortal engine games!
You've made a very good point about the engines not knowing the best moves either.(at least in the early midgame) It's so jarring to hear commentators treating the engine moves as if chess were solved & those were the best possible moves in existence.
The opening is just unreal.... astonishing game 😳
The 50 games black/white is for the TCEC SuperFinal. This is from the Cup, that starts off 'best of 4'
Leela won this one 3-1 btw.
knight sacrifice at the beginning to trap black pieces is unthinkable
This is nuts!
@agadmator I would really appreciate the comparative analysis between different modern systems.
Please do it :)
Would love to see the different fundamental ideas in engine intrinsics. Maybe in a future video?
Wow Leela is some powerful engine!! Stifled the other engine and it couldn’t breathe🙆🏽♂️
Didn't know I needed to hear agad philosophize about chess engines today
Yes but at move 57, Leela played Qc7+ and then 6 more moves, but she had Qd8 which needed just 5 more moves till checkmate. It's possible that she doesn't care of checkmating faster after all, as long as a checkmate has been seen.
Engine shows no mercy
Moooooore of these