Leela remarkably discovers theoretically critical variation against Modern Benoni vs Stockfish

Рет қаралды 9,683

5 жыл бұрын

♚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 KZfaq Chess Video Channel : bit.ly/kingscrusher
🎥 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
♚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

Пікірлер: 53

@kingscrusher 5 жыл бұрын

Q. Can Neural Networks discover from scratch the most theoretically important openings ? 🤔😀😎🌍 Video: goo.gl/koGHtn Replayable game: goo.gl/r8vyD3 #chess #chessgame #ArtificialIntelligence #AI #MachineLearning #DeepLearning #chessopenings

@tomasseeber 5 жыл бұрын

Leela kept moving the queen, because SHE DOESN'T WANT TO DIE by the game ending.

@inversehyperbolictangent3955 5 жыл бұрын

Lol. 😄 Paraphrasing another great (albeit fictional) game-playing AI: "A STRANGE GAME. THE ONLY WINNING MOVE IS NOT TO STOP PLAYING."

@tomasseeber 5 жыл бұрын

@@inversehyperbolictangent3955 Gigantic comment!

@jamesecarson5631 5 жыл бұрын

"perfect for a knight popping in for dinner." "the diagonal of death". "it's going to be a world of pain". "Stockfish plays h6, which may be the best of the bunch". "take out e3 at leisure".

@balthazarbeutelwolf9097 5 жыл бұрын

The Leela test30 nets are known to delay mating in the endgame till the 50 rule forces them - in TCEC that does not show normally because of adjudication, but in CCCC.... Actually Komodo MCTS has the same issue, therefore that engine switches to AB mode when it reaches a clear (+7 or so, do not remember) advantage.

@moosewild4239 4 жыл бұрын

that queen at the end too funny leela showing human emotion rubbing salt in the wounds like a cat playing with its prey

@inversehyperbolictangent3955 5 жыл бұрын

Wow, interesting analysis and history lesson there. Thanks, KC! 😊

@pazdziochowaty 5 жыл бұрын

If Leela really used tablebases she would win quicker. K+Q vs K from any position is at most mate in 8. If she uses TB and still goes for mate in 50 then devs should work on improving TB module otherwise there is little difference from Leela making the moves on her own

@johntate6537 5 жыл бұрын

When you think about it, there is as logic to why Leela might be taking so long to close out winning positions. To us it seems obvious that it is advantageous to win quickly - to save energy if for no other reason. My guess though is that this kind of timing issue has never been included in Leela's training regime. To her a 350 move game is as good as a 50 move game; the winning is all that counts.

@Th3Wick3dOn3YT 5 жыл бұрын

KC, I dont think you understand, you know kids when they ride bikes and they go " Look MA! No hands!" Thats leela's Ba4+ , " look guys I can beat stockfish without the bishop!" The logic is simple, without the bishop I have less moves to consider! LOL.

@piyushsharma5938 5 жыл бұрын

Leela proves highly intelligent people always have been a bit eccentric. .. I heard even Newton made two holes in his gate. .. A bigger hole for the bigger cat and another smaller hole for the kitten. ..

@kingscrusher 5 жыл бұрын

Fun comment - Cheers, K

@bhgtree 5 жыл бұрын

piyush sharma Ha ha, I completely agree, its as if Leela wanted to be different from the rest :)

@chesslover2925 5 жыл бұрын

Tx KC. This game made me laugh a lot. Amazing and funny

@modolief 5 жыл бұрын

Great game, thanks!!

@peterpetrov6522 5 жыл бұрын

Very entertaining video!

@MrSilvershane 5 жыл бұрын

Thanks KC. I wonder what the most moves played in an engine match is?

@julioandresgomez3201 5 жыл бұрын

The troll strikes again.

@XEJTEP Жыл бұрын

I saw few positions where you can make fortress with bishop and knight or 2 bishops ( different color ) against queen and get a draw . If someone could pull that would be amazing .

@jbrassesco 5 жыл бұрын

I must be missing something. On 17:31 we reached this simple position: White: Kg8, Rg7 and pawn in h7. Black: Ka4, Bb2 and knight in f5. And white played pawn queens in h8??????? What about Ra7+, Kb3 and Rb7+, taking the bishop on the next move? Queen vs knight instead of Quenn vs knight and bishop

@jarirepo1172 5 жыл бұрын

Leela does not care about winning most efficiently, just winning. Sometimes it seems like the more the moves it takes, the more she likes that line.

@jbrassesco 5 жыл бұрын

@@jarirepo1172 Yes, it must be that. Same thing happens in Go with AlphaGo

@doncar9 5 жыл бұрын

Thanks Tryfon,that was a great game and I laughed like hell at the last 250 moves! Thank you for showing computer games . They are so much more interesting than human ones ,particularly in the Carlson v Caruana match ? Why do you think that is?

@Bobby-fj8mk 5 жыл бұрын

Thanks KC - have you got any games where Leela crushes Stockfish in a middle game with a forced checkmate?

@silas4853 4 жыл бұрын

You should have added each opponent name and their engines strength rating.

@folwr3653 5 жыл бұрын

Thanks KC.

@kingscrusher 5 жыл бұрын

Cheers, K

@Marcusjnmc 4 жыл бұрын

was stockfish going e6 rather than e5 the losing move here allowing Leela's brilliant later f4 ?

@Marcusjnmc 4 жыл бұрын

oh '.' it was still in book at that point, so it was forced to

@rogerlie4176 5 жыл бұрын

I checked the position at 16:06 with 7-men tablebase, and it turns out that white could win one of black's pieces in 27 moves.

@fuckgoogle5277 5 жыл бұрын

At 2:04 he said white is up in material but by my count black has more material right?

@evianwahter 5 жыл бұрын

Fuck Google 2 piece for rook

@pnachtwey 5 жыл бұрын

I don’t think these opening are truly tested with such short time controls.

@Wight054 5 жыл бұрын

Hey Kingcrusher could you post the pgn to this game please?

@kingscrusher 5 жыл бұрын

Full members or channel members here can export annotated PGN from this link on login: www.chessworld.net/chessclubs/ltpgnviewer32/ltpgnboard.asp?GameID=5024128

@Wight054 5 жыл бұрын

Thank you kind sir!

@douglasquaid7550 5 жыл бұрын

"Diagonal of death"

@paulbloemen7256 5 жыл бұрын

Winning a won game, from move 100 onwards. What a flexible mind does one need to have, after such an exciting opening and middle game, to play such a super dry endgame. White has a decisive advantage, but it would take some 150 horrifying moves to get the win under your belt. If AI-Zero is going to be of huge significance outside the games, it would seem desirable to me to build in some notion of efficiency, which is not conflicting with the Zero principle. In chess it is speed, in go it is maximum winning margin. Just some expectation guessing: within 2 years?

@jarirepo1172 5 жыл бұрын

Yeah that trolling part is quite funny and strange at the same time, seems like it's enough for Leela to find a win, no matter how unefficient. At the opening and middle game moves must be extremely precise so no such problem there, of course...

@paulbloemen7256 5 жыл бұрын

Jari Repo I wonder whether Leela, AI-Zero in general, has a notion about the difference between opening, middle game and endgame. I think it only sees "position" and runs its algorithm from there, finishing all the evaluation games to pick the move with the highest chance to win the game. But apart from the chance to win the game it seems to have no additional sub-criteria like efficiency. Adding them, and maybe other ones, could be necessary to make AI-Zero all the more useful for our human needs: we humans are good at striving for different, sometimes conflicting objectives, in the end AI-Zero must provide here too.

@jarirepo1172 5 жыл бұрын

@@paulbloemen7256 That's what I think as well, it's just that earlier in game usually there are a lot bad moves and few good ones; later you can (and Leela will) shuffle a lot if your opponent has no counterplay left. Maybe time format affects this as well, something like Leela making moves as fast as possible to conserve time and only spending some when 50-move rule is threatening or something?

@franklippert4278 5 жыл бұрын

Strange, Leela more or less crushes SF but then can't (won't) mate with K+Q vs K? Somewhere the Leela developers must make changes to the way the reinforcement learning is applied.

@evianwahter 5 жыл бұрын

A tip for people on making money lmao. I was told that it takes 7 exposures before you are ready to buy something and right now I'm sure I have watched 700 ads from tai Lopez and it's getting hard not to spend money yo. So that's all you need "drive" and consistency

@user-dl8vd8tv3d 5 жыл бұрын

Hi man Black con play and draws with fortress

@user-dl8vd8tv3d 2 жыл бұрын

am from future and be7 was a mistake

@user-dl8vd8tv3d 2 жыл бұрын

best was b6

@bagoplayer7455 5 жыл бұрын

which network?

@Askhat08 5 жыл бұрын

32425

@victorsamuelbaezsouffront9017 5 жыл бұрын

do you says crushing every time that you speaks :V? damm thats why kingcrusher :V.

@mojohojo4077 5 жыл бұрын

50 move draw???

@NickTF2 5 жыл бұрын

another terrible bishop

@jarirepo1172 5 жыл бұрын

One of the things I would look into if I played this as black is how to get anything done with that light square bishop, it was just useless when it mattered... even worse, just creating tactical chances for d-pawn push and losing that bishop for a pawn.