Highly Evolved Leela Chess proves beyond doubt beautifully liberating pawn sacrifices vs Stockfish

♚COURSES kingscrusher.tv/chesscourses
📚 kingscrusher.tv/chessopenings
📚 kingscrusher.tv/chesstactics || kingscrusher.tv/londonsystem
#KCComputerChess
#KCLeelaChess
♚ Play turn style chess at bit.ly/chessworld
♚ Play Chess vs. Kingscrusher and others: www.chessworld.net/chessclubs...
♚ Subscribe to best KZbin Chess Video Channel : bit.ly/kingscrusher
🎥 Immortal Must See Chess Games: bit.ly/2KiWGhP
🎥 Leela Chess Playlist: bit.ly/leelachess
FIDE CM Kingscrusher goes over amazing games of Chess every day, with a focus recently on games of Neural Networks which are opening up new concepts for how chess could be played more effectively.
👕 Kingscrusher Store: teespring.com/stores/kingscru...
♚Donating via Paypal: goo.gl/7HJcDq
🎥 Immortal Must See Chess Games: bit.ly/2KiWGhP
♚Channel exclusive perks: / @kingscrusher
Game quality tags: amazing, awesome, astonishing, brilliant, classic, crushing, dynamic, elegant, exceptional, excellent, exciting, fabulous, famous, fantastic, finest, flashy, greatest, important, impressive, incredible, instructive, interesting, magnificent, marvellous.
Info about Leela Zero:
en.wikipedia.org/wiki/Leela_Zero
...
Leela Chess Zero (lc0) is a free, open-source, and neural network-based chess engine and distributed computing project.
Leela Zero's algorithm is based on DeepMind's 2017 paper about AlphaGo Zero.[3][6] Unlike the original Leela, which has a lot of human knowledge and heuristics programmed into it, Leela Zero only knows the basic rules and nothing more.[7]
Leela Zero is trained by a distributed effort, which is coordinated at the Leela Zero website. Members of the community provide computing resources by running the client, which generates self-play games and submits them to the server. The self-play games are used to train newer networks. Generally, over 500 clients have connected to the server to contribute resources.[7] The community has provided high quality code contributions as well.[7]
Leela Zero finished third at the BerryGenomics Cup World AI Go Tournament in Fuzhou, Fujian, China on 28 April 2018.[8]
Info about Alphazero:
en.wikipedia.org/wiki/AlphaZero
AlphaZero is a computer program developed by the Alphabet-owned AI research company DeepMind, which uses an approach similar to AlphaGo Zero's to master not just Go, but also chess and shogi. On December 5, 2017 the DeepMind team released a preprint introducing AlphaZero, which, within 24 hours, achieved a superhuman level of play in these three games by defeating world-champion programs, Stockfish, elmo, and the 3-day version of AlphaGo Zero, in each case making use of custom tensor processing units (TPUs) that the Google programs were optimized to make use of.[1] AlphaZero was trained solely via "self-play" using 5,000 first-generation TPUs to generate the games and 64 second-generation TPUs to train the neural networks, all in parallel, with no access to opening books or endgame tables. After just four hours of training, DeepMind estimated AlphaZero was playing at a higher Elo rating than Stockfish; after 9 hours of training, the algorithm decisively defeated Stockfish 8 in a time-controlled 100-game tournament (28 wins, 0 losses, and 72 draws).[1][2][3] The trained algorithm played on a single machine with four TPUs.
...
Relation to AlphaGo Zero
Further information: AlphaGo Zero
AlphaZero (AZ) is a more generalized variant of the AlphaGo Zero (AGZ) algorithm, and is able to play shogi and chess as well as Go. Differences between AZ and AGZ include:[1]
AZ has hard-coded rules for setting search hyperparameters.
The neural network is now updated continually.
Go (unlike Chess) is symmetric under certain reflections and rotations; AlphaGo Zero was programmed to take advantage of these symmetries. AlphaZero is not.
Chess can end in a draw unlike Go; therefore AlphaZero can take into account the possibility of a drawn game.
AlphaZero vs. Stockfish and elmo
...
►If you like my chess videos please consider become a full member here at my Chessworld Site. www.chessworld.net/chessclubs/...
#KCChess game tactics
♞ Challenge KC and others for turn style chess at bit.ly/chessworld
♚TEAM kingscrusher.tv/team
♚TWITTER kingscrusher.tv/twitter
♚DISCORD kingscrusher.tv/discord
♚TWITCH / kingscrusher
♚REDDIT kingscrusher.tv/reddit
♚QUORA kingscrusher.tv/quora
♚DONATE paypal.me/kingscrusher
♚COURSES kingscrusher.tv/chesscourses
♚COURSES kingscrusher.tv/chesscourses

[@kingscrusher]

Q. Can Neural network Leela Chess break positional binds ? 🤔😀😎🌍 Video: goo.gl/cLzSC3 Replayable game: goo.gl/TC1upK #chess #chessgame #ArtificialIntelligence #AI #MachineLearning #DeepLearning

[@FarshadTorkashvand]

only Leela trolls Stockfish. lol

[@kellmano1]

Leela just won a mad game (204) on chess.com. Sacrificed 2 rooks for a Bishop and a couple of very strong pawns. no idea how correct it was

[@julioandresgomez3201]

Leela toys around even with a second on the clock.

[@kingscrusher]

Cheers, K

[@rdecredico]

One second to this brain is like several hours for a human being.

[@4merxtian432]

End of AB engine era soon. I imagine a hybrid SF Leela engine would be insane and not too difficult to develop

[@kingscrusher]

Fun comment - Cheers, K

[@pazdziochowaty]

Is this really same network playing in the whole game? Seems like a 3500 version built the winning position and then a 800 elo beginner took the mouse over and had no clue of how to checkmate the lone king. Does Leela have a younger brother and let him finish her games? :)

[@dubsphinx]

Great commentary, very appreciated. I am loving the Leela games, keep them coming king.

[@kingscrusher]

Cheers, K

[@CarloRossiTheGreat]

Thanks for the upload and analysis. Can't wait to get into this

[@kingscrusher]

Cheers, K

[@GurukulStudyCentreIndia]

Thank you 🔔🎺 great analysis.

[@kingscrusher]

Cheers, K

[@driewiel]

Checkmate in 600 moves! I think Leela likes playing chess.

[@kingscrusher]

Fun comment - Cheers, K

[@edwardsolomon1951]

My only logical explanation for Leela supposed "trolling" at end is that Leela stalemated itself in self-learning games when it had more pieces on the board, therefore it's end-game "knowledge" is grotesquely warped causing it to sacrifice and underpromote to avoid potential stalemates, whereas only having a K+R or K+ Q has vastly LESS chances of stalemating. My 2 cents. And come on, how many of us delivered stalemates in otherwise winning positions in Blitz chess? All of us lol.

[@brendanbaker7111]

Edward Solomon I think it's just she will play any move with 100 percent win chance so if she has let's say a queen and a king vs just a king almost all moves evaluate too 100 percent then she randomly chooses any move that won't draw since she has no preferace for faster mates all moves with the same expected win percentage are equal in leela mind.

[@quag443]

The trolling has nothing to do with the neural network itself. Rather, it's a result of sloppy tablebase implementation. Leela has been programmed to *randomly* play a winning move rather than play the *quickest* winning move. It's actually been fixed recently ( github.com/LeelaChessZero/lc0/pull/708 ) but it has yet to be be merged to the master branch. Hopefully it will in the next few days!

[@edwardsolomon1951]

@@quag443 From what I k ow about Leela is that it isn't programmed at all beyond the rules of the game that are required for it play against itself in order self learn and program itself.

[@truthbetold1377]

"snug as a bug in a rug"

[@michaelmorris4515]

Leela's trolling is getting worse. It's as if she has "make the game go as long as possible" as a secondary goal.

[@mtgradwell]

I think you speak maybe more truly than you realise. "Make the game go as long as possible" is indeed her secondary goal. it's because of the way tablebase integration was implemented. In circumstances where throwing away a queen would result in a tablebase win, Leela's weights were adjusted to make throwing away the queen preferable to keeping it. This was even done retrospectively on training games that Leela had played before the tablebase was integrated. But Leela did not and still does not know anything about tablebases. All she knows is what the weights tell her - that in certain unspecified circumstances, if the ultimate win is still assured, it is preferable to throw stuff away and go for a narrow and long-drawn-out victory than it is to go in for the immediate kill.

[@matteogauthier7750]

7:58 A mate is "very pleasant" for black? I mean, you're not wrong...

[@aaronsmith6632]

This is a premonition of just how sadistic our future AI overlords will be when there's information they would like to extract.

[@rdecredico]

2 seconds is an eternity to these brains.

[@kingscrusher]

Fun comment - Cheers, K

[@paulbloemen7256]

Interesting! I tried to figure out which was the last move where White could have gained a slight advantage. Is 7. Qc2 the losing move, where 7. Nc3 would have been better for White? It would be quite something if a move like 7. Qc2 would be the losing move. As part of the opening book.

[@ddstar]

17:17 BM... LOL

[@ew6230]

A great ruler knows that loss is inevitable and minimizes loss as well as maximizing gains!

[@kingscrusher]

Cheers, K

[@peterpetrov6522]

17:17 LOL!

[@kingscrusher]

Fun comment - Cheers, K

[@IMortage]

Will you take a look at the new AlphaZero material in GM Matthew Stadler's new book?

[@alexmagor7538]

Seems like Leela likes to torture her opponents in the end

[@kingscrusher]

Fun comment - Cheers, K

[@plasmaastronaut]

seems like a game that will make stockfish programmers lose their hair. giving away central pawns, allowing an opponent knight to mobilize into the centre, and losing a tempo to move the queen. how they are going to code this into SF11 I don't know. maybe chess cannot be tamed by rules, there will always be scope for mad moves that win. leela delivers another superhuman masterpiece

[@shanu9008]

Sir Why always stockfish use another engine against alfazero

[@UKmarkj]

Something a little eerie about Leela trolling. Has the AI been programmed to do this or why does she do it...? :O

[@mtgradwell]

it's a consequence of the way tablebase was integrated. Weights were tweaked to tell Leela that a tablebase win was preferable to anything, including massive material superiority. This might seem to make sense superficially, given that a tablebase win is guaranteed; but Leela didn't and still doesn't know anything about tablebases, doesn't even know that they exist. All she knows is that in some circumstances, generally where there aren't many pieces left, throwing pieces way for no obvious reason is somehow a good idea. Even throwing away a queen is a good idea. It's because the position without the queen is in the tablebase, and the position with the queen isn't, but Leela doesn't know that. I think that in trying to make sense of this Leela has concluded that if the win is guaranteed then a narrow victory is preferable to a decisive one, and a long game is preferable to a short one.

[@qwertz12345654321]

Can someone explain why leela loves to make redundant and stupid moves when the win is guaranteed? Isn't there a incentive to win faster if there is a forced win?

[@-_Nuke_-]

Propably trolling developers xΔ

[@bhgtree]

The NN does not try to win quickest, when Leela is in a position that, for example, 10 different moves all equally wins, she doe not try to find the shortest route.

[@Askhat08]

She cannot distinquish different moves then because all of them are winning. Only when 50-move rule hits, she starts moving.

[@anythingpeteives]

Time (and amount of moves) has no concept to a chess engine. A win is a win, is a win. And the safest possible win is when there is just a King and a Queen against a King as far as Leela is concerned.

[@qwertz12345654321]

@@anythingpeteives maybe not for leela, but pretty much any engine prefers faster wins

[@Nuhyamin1]

Leela!

[@kingscrusher]

Cheers, K

[@greenmindai_mihael]

Why Leela can not give fast check mat. What is with this sadistic end games with +15?

[@piyushsharma5938]

Lesson learned. .... never to get annoyed by a sadistic human. ... coz even machines with A.I. are sadistic. .... maybe sadism is ok.... and there was a fault in me who always opposed sadistic behaviour

[@kingscrusher]

Fun comment - Cheers, K

[@LuisManuelLealDias]

Jesus Christ do they need to train better "shuffling" mechanics. That ending is just ridiculous!

[@yggdrasil9039]

Whites moves around 15:43 remind me of those electronic chessboards from the 1980s that made the dumbest moves when they start losing. Why doesn't white at least put pawns on white squares, why move them to dark squares? That makes zero sense.

[@RalfStephan]

Again and again, Leela wins OCB endgames. Makes you wonder.

[@kingscrusher]

Great comment, K