Google Deep Mind Alpha Zero Sacs a Piece Without "Thinking" Twice

Download Mproov and Improve Your Chess Today! app.mproov.me/...
Follow MprooV on Twitter / mproovapp #agadmator Read more about Deep Mind here arxiv.org/pdf/...
Link to the other games lichess.org/st...
A chess game between Deep Mind and Stockfish
Google Deep Mind Alpha Zero vs Stockfish
One of the games
1. Nf3 Nf6 2. d4 e6 3. c4 b6 4. g3 Bb7 5. Bg2 Be7 6. O-O O-O 7. d5 exd5 8. Nh4 c6 9. cxd5 Nxd5 10. Nf5 Nc7 11. e4 d5 12. exd5 Nxd5 13. Nc3 Nxc3 14. Qg4 g6 15. Nh6+ Kg7 16. bxc3 Bc8 17. Qf4 Qd6 18. Qa4 g5 19. Re1 Kxh6 20. h4 f6 21. Be3 Bf5 22. Rad1 Qa3 23. Qc4 b5 24. hxg5+ fxg5 25. Qh4+ Kg6 26. Qh1 Kg7 27. Be4 Bg6 28. Bxg6 hxg6 29. Qh3 Bf6 30. Kg2 Qxa2 31. Rh1 Qg8 32. c4 Re8 33. Bd4 Bxd4 34. Rxd4 Rd8 35. Rxd8 Qxd8 36. Qe6 Nd7 37. Rd1 Nc5 38. Rxd8 Nxe6 39. Rxa8 Kf6 40. cxb5 cxb5 41. Kf3 Nd4+ 42. Ke4 Nc6 43. Rc8 Ne7 44. Rb8 Nf5 45. g4 Nh6 46. f3 Nf7 47. Ra8 Nd6+ 48. Kd5 Nc4 49. Rxa7 Ne3+ 50. Ke4 Nc4 51. Ra6+ Kg7 52. Rc6 Kf7 53. Rc5 Ke6 54. Rxg5 Kf6 55. Rc5 g5 56. Kd4
----------------------------------------------------------------------------------------------------------------------------------------
If you realllly enjoy my content, you're welcome to support me and my channel with a small donation via PayPal or Crypto.
Link to PayPal donation www.paypal.me/...
Maiar Wallet @agadmator or get.maiar.com/...
BTC address bc1qckd3ut0hqyymzv33eus97ld8klj02xhk2kcwld
BCH address qzmfclyn69hqhjslls40r7r0dsttwe3tcsl946w4fr
LTC address Laarf1RmvCpLt2BcSwC1PBLG3hRC4HjBrz
NANO nano_1h1kgfaq88t1btwadqzx73rbha5hwbb88sxmfns851kwj8hnosdj51w388xx
Monero 4AdvvqmC4xhPyyRSAEDxTTAoXdxAtX2Py6b8Eh4EQzBLGbgo5rY5Khcap1x76JrDJH87yibAE9b6TPwTsvBAiFFCLtM8Be7
For any other currency address, contact me via [e-mail]
Check out some of the books I enjoy www.amazon.com...
Check out ALL my videos here • You Snooze, You Lose |...
Facebook: / agadmatoryoutube
Twitter: / agadmator
Instagram: / agadmator
Lichess: lichess.org/@/...
Chess.com: agadmator
League of Legends: agadmator (EUNE, my friend is using my EUWE account for a couple of years now)
Blizzard: agadmator #2992
Join our Discord Server here / discord
p.s. if you work in Twitter or Instagram, help me get Verified :) "Watch me without ads on your Amazon devices (bit.ly/Agadmat...) and Roku TV (bit.ly/Agadmat...)

[@adflicto1]

8:36 It's very satisfying to hear the sentence: "Stockfish resigned the game" Haahhahahahahah that was amazingly funny

[@MrSupernova111]

hahaha

[@RyanSmith-ow6cm]

As a go player who wandered into these videos, that's what it sounded like reading that AlphaGo Zero beat AlphaGo Lee 100-0. Doesn't sound real does it? :)

[@ThePCBetaphish]

Aleksandar Spasov I was thinking the exact same thing. It's the funniest thing.

[@abebuckingham8198]

In the methodology it mentions that Stockfish resigned only if it evaluated the position as -9 or worse for 10 consecutive moves. It was certainly convinced that the position was lost in the end.

[@blugaledoh2669]

Stockfish is coward, I will fight to the end.

[@L2Pmugmouse]

I understand about 3% of what this guy talks about. I don't even play chess. But I can't stop watching these videos.

[@notjoke6164]

shawn hahhahaha

[@josedavidbenavides3922]

Hahahhha

[@dikshantraj6005]

Same

[@zmlgboss8479]

Same

[@xJopeRz]

HAHAHAHA same wtf

[@dariobarisic3502]

Karma to stockfish for defeating me countless times.

[@ekinbasken8891]

Dario Barišić yeah take that stockfish

[@timothysiahaan8977]

Lol!

[@Koolduderavi]

Haha

[@raviravi5342]

lol

[@krishnapidaparthi6489]

Dario Barišić innovative comment

[@travisheck5979]

Dude what is amazing to me is that alphazero is almost playing like its in the romantic era of chess with its daring sacs and gung ho attacks, a style that fell out of favor a long time ago. Seems like its playing with an eerie amount of intuition and positional play, tossing away pieces to gain overwhelming control of the board

[@DividendsMaster]

Travis Heck truuuue

[@Mastouril]

deepmind just digs chess

[@takatotakasui8307]

It makes sense that he would have an older style, because he trained himself. He doesn't know anything about chess theory, he just finds the fastest way to win.

[@xeofreestyler]

Pankaj R its play reminds me of Mir Sultan Kahn

[@Jonathan-ec9pp]

More like Morphy, I'd say

[@drawsgaming7094]

*Stockfish has left the game*

[@RedLunarFox]

Nice 👌

[@Wick291]

Draws Gaming Stockfish plays Queens Indian defense against AlphaZero... *Its not very effective*

[@tarasov9794]

*chat

[@gluonpa6878]

SF ragequit

[@xyon9090]

*AlphaZero:* Here eat my pieces *Stockfish:* Omnomnom *AlphaZero:* Yeah that's right. Become a big nice fish for me to catch later on

[@jkgan4952]

Lol

[@saileshraut7151]

Omnomnom lol

[@ichbinrudi]

Best comment on youtube (i'm dead serious)

[@rajithasan9319]

@@ichbinrudi rip dear :')

[@heldenhonig6505]

This sounds awfully sexual

[@rcstockton]

"In this position, Stockfish resigned the game." Glorious.

[@Kinjutsuu]

"Maybe it can even teach Max Deutsch to beat Carlsen in a month" ROFLMAO!!!! instant subscribe and liked, freaking love you for saying that :D

[@agadmator]

+Liszt Well, it might :D

[@hey8174]

Now all the chess players know what Ke Jie was feeling when he said Alpha Go played like a god. "I'll leave him with the task to explore the universe while I, myself, prefer fishing in the pond."

[@Kinjutsuu]

Still, the best Go player managed to beat it once. It shows that chess is a much easier game for engines

[@ballom29]

but that was against alphago lee alphago lee < alphago master < alphago ke jie

[@Joellercoaster]

Not the Zero version. The version of AlphaGo that was defeated once by Lee Sedol was substantially different to the engine that is now being used to learn chess... the Zero (meaning "zero human strategic programming") approach taught itself go and crushed previous versions of AlphaGo (including one that went undefeated against all human players, learning from the defeat by Lee). This is not saying anything about chess vs go, but rather about the current generation of DeepMind's technology vs anything that came before.

[@jeanconan1812]

@@Kinjutsuu well, fairy chess like terachess are actually more complicated than go.

[@BeerdyBruceLeeCentral]

More computer games please.

[@heffalump111]

I concur--why does he not like to show beautiful A.I. chess?

[@billschannel1116]

Agadmator knows that if he covers too many Ai games, the AI's might learn what makes a successful chess analysis channel.

[@CypherOzzie]

NOOOOOOOOOOOOO!!! BORING! People are much more fun

[@timangar9771]

Beerdy - Bruce Lee Central hi beerdy, do you think bruce liked chess?

[@BeerdyBruceLeeCentral]

He did play chess, but he had no time to get any good. He spent 10 hours a day training martial arts. 7 days a week. No time for chess.

[@nathanjxaxson]

It's pretty amazing how positional Deep Mind plays. It is just a brutal attacking monster.

[@gmsi7d371]

yes this is insane calculation. even carlsen would be destroyed quickly

[@areyouscaredyet6916]

@@gmsi7d371 two decades ago, computer engines were already stronger than the top players. Magnus has stated he doesn't play the computer anymore because he says he doesn't like losing to an "idiot". These engines make no sense in human perspective, but they are simply too strong for Magnus to ever learn anything from watching it play.

[@vyshawnforeman8384]

@@gmsi7d371 of magnus carlsen would get destroyed no contest

[@mme9646]

I agree with Carlsen, there's just no sense in playing against machines. It's like comparing the world's strongest weightlifter to a forklift. Just doesn't make sense

[@Abhishek-150]

Top chess players have good memory and many of new players remembered the engine moves by playing with them multiple times on many many different positions.. They became very strong but also very stupid at the same time.. This Alpha Zero is a wake up call for those players..

[@samkachar]

Tears to my eyes when I heard, this next sentence gives me great pleasure to say. Stock fish resigned the game. That's a man with some hate in his heart toward that engine.

[@Manikese]

“It taught itself chess in 4 hours, I’m pretty sure it will be a valuable member of the chess community.” 😂

[@bardhanjoy]

Deep mind didn't play like an "engine", it played like a supremely skilled human chess player. It's not materialistic. It's goal oriented. Not "boring" at all rather it's problem solving method is like a saint or an assassin. Very goal oriented but the problem solving manner is brilliant yet relatable to humans :)

[@Wyvern07_]

Joydeep Bardhan beautifully said

[@bardhanjoy]

It was such a game :)

[@huuquyennguyen348]

Exactly what I think. It played so human

[@kher0]

My thoughts exactly It was very very very entertaining.. way more entertaining that most of grandmaster matches I have watched

[@Gommerell]

@@huuquyennguyen348 It played so Human that no human can beat it.

[@villaholland]

AlphaZero gonna test openings. Also I wonder if AlphaZero can solve locked fortress positions that normal engines struggle with.

[@dmtree__]

Alpha Zero seems to value initiative much more than 'brute force' engines. This might, indeed, change the way we evaluate positions. Exciting

[@harbhub]

Might? It's definitely going to change how we evaluate positions.

[@ritsukasa]

not easy, its very well done, at 3500 elo level

[@santhoshrevuluri4187]

Mikhail Tal will be smiling from heaven

[@harabas3499]

together with Spielmann, Keres , Euwe and Nezhmetdinov

[@persiankingish]

😂😂😂😂

[@kasparov937]

Santhosh Revuluri Hey dont forget Kasparov!!!

[@dasgafs6297]

He's not dead yet tho...

[@GBValode]

There is no heaven , there is no hell. I thought that chess players are pretty smart in general...

[@ImaginaryHuman072889]

amazing. alpha zero is like morphy in the opening (breaking alot of opening theory "guidelines"), tal in the middle game (sacrifices pieces to gain initiative), and capablanca in the endgame (just very calmly and cleanly finishing up the game)

[@aracaniusinfinius2880]

thanks to yours and suren's channel, i have been able to improve my level of play in terms of tactics from a measly 1100 after having not played for about a total of 4-5 years not having practised to a tactics rating of about 1500 in about 2 months of training :D

[@JuvoII]

I'm gonna give it 4 hours, and we shall see if I have a chess future.

[@giovannimounir4744]

Please where I can get this rating??

[@abdualziz9]

@Giovanni Mounir chess.com or lichess.com

[@giovannimounir4744]

Abdualziz thanks

[@jyashin]

I'll have to assume this is the old Elo rating, and not Glicko. 1500 Glicko = 1200 Elo.

[@teo5146]

"Alpha Zero's win was not smooth enough" - Maurice Ashley.

[@agadmator]

+aDayinUruguay :)

[@9181shreyasbhatt]

why not make Maurice Ashley play AlphaZero

[@MrSupernova111]

Hahaha

[@bradc3402]

Because it would be like taking candy from a baby.

[@Scy]

It would be like taking candy from a baby, spiking it with cyanide and then shoving it up its arse.

[@haricharanbalasundaram3124]

Wow... Love your videos, man, continue this!! Very awesome game especially against a stronger stockfish!!

[@bra-steve]

It seems that making a lot of Queen moves is not that bad of a thing😏

[@alhassanali4829]

Only when you know what you're doing

[@ln9296]

wrong

[@skddwivedi2158]

If you're doing it with tempo.

[@AK-rx6hv]

Few things I learned from alpha zero: -Nothing wrong with moving pieces right back to where they were a move ago (That's otherwise usually considered a wasted move) -No need to link the rooks -Pawns are sacrificial lamb -Trade Queens -Slow and steady wins the race, wear the opponent down, no need to attack early

[@NJ-wb1cz]

I think all of that makes sense if you're Alpha Zero. The rules humans have reflect limits humans hame

[@kusy2130]

Show Deep Mind Alpha Zero as black.

[@macnolds4145]

Such a "pure" piece sacrifice. There was no direct material compensation or immediate checkmate...AND Black was up a few pawns! Yet, AlphaZero correctly assessed that his positional superiority and the undeveloped Black pieces would allow him to convert his initiative into a dominant win.

[@krishnapidaparthi6489]

After all, the very innovative dialogue in this video is "HOW TO BEAT CARLSEN IN ONE MONTH" @Agadamator

[@jayk8772]

Amazing... if you set up Stockfish to play itself, it saves the knight every time but draws in the end.... amazing this Alpha Zero created an imbalance that allowed it enough play to create an advantage

[@anthonygross1963]

Qh1! move of the year.

[@FreethinkingSecularist]

It won as black 3 times against Stockfish. Those are the games that might really be interesting from you! #Requestgame I forgot what hashtag you said to use for suggestions. #suggestion

[@FreethinkingSecularist]

No DeepMind Alpha zero did.

[@FreethinkingSecularist]

Even God draws 75% against Stockfish 8.

[@naj40]

Lmao same I read it as "I won as black 3 times against stockfish"

[@jamesfleming1155]

It seems to win these games not by using advanced tactics, but by outplaying positionally. You would think humans would be able to play at this level. I always assumed humans lagged behind computers because we can’t analyze as many lines. But I don’t know if that’s what deep mind is doing.

[@ballom29]

80 000/sec? maybe not,but you probably calculate a lot/sec, it's just you determine immediately if they're bad, letting yourself thinkign longer for moves that sound good.

[@InXLsisDeo]

Neural networks "thinking" is hierarchical in nature: they "see" global shapes/schemas/patterns, and then go deeper and deeper in the specialization with each layer of neurons until the general shape reduces to one answer. In this way, it's very much like human intuition. That's why go was so adequate for A0 and search tree algorithms aren't when the search tree is huge. That also may explain why A0 is so strong at positional and long term "thinking". So A0 reasons a bit like us, but it has the advantage of never making mistakes in the process. The human brain has many many more neurons and many many more connexions, but they don't work as fast, they do a lot more tasks (interact with the world, manipulate language, regulate body functions, etc) and more importantly, they can fail.

[@ekki1993]

Humans lag behind computers both because of the amount of data they can process in a given amount of time (80k positions per second is impossible for a human) and because humans make mistakes much more often than computers. At least for chess engines like stockfish, their "insight" into chess was built-in by humans, so it was no greater than any human's insight into the game. Alpha Zero, on the other hand, made its own "rules", so there's no way (yet) to know if it's a set of rules achieveable to human intellect or if it's unachieveable without machines.

[@erikbartalos830]

It does not matter AlphaZero had 'only' 4 hours, it was run on 4000 TPUs, which is INSANE from the pure computing capacity.

[@alexdavn6093]

it was trained on 5000 TPUs, it was run on 5 TPUs big difference Stockfish was run on 64 CPUs and "trained" with thousands of Cpu-hours and countless man-hours so, fair contest

[@erikbartalos830]

not really, as TPU is far superior than CPU for these kind of tasks

[@alexdavn6093]

Well, it searches only 80k positions per second compared with 70 million for Stockfish during play, so the magic has to happen in the training phase. Just speculation but lets say a TPU is 10 times more efficient as training on a top tier GPU (seems reasonable) that would be 50 000 *4 hours(time necessary to get as strong as stockfish) = 200 000 hours divide by 64 cluster of GPUs (comparable to 64 CPUs for Stockfish) 3125 hours so 130 days Stockfish needed a lot longer and lots of human input to get to this stage. ps. obviously i simplified and speculated still

[@InXLsisDeo]

Erik Bartalos Not at all, a TPU's task is to do calculations on tensors and matrices (the mathematical objects). The T stands for Tensor. Nothing to do with chess or go. It's like saying a graphic card is far superior to a CPU for playing chess. If you want to see a good introduction to how neural networks work, go watch the videos from 2Brown1Blue on the subject.

[@garylake8654]

Using the queen to gain the tempo, then keep said tempo up with a sacrifice, thereby rendering Stockfish's A & B major pieces as spectators, is just so intuitive, its like an watching an artist playing a machine, akin to observing a reckless version of myself, but with a very different outcome....

[@aliakbarpoonawala6168]

Deep Mind Alpha Zero is the next world champion

[@thor9695]

he already is lol

[@olefster]

Deep Mind Alpha Zero is the next overlord. Call to John Connor. Immediately

[@columbus8myhw]

They need to enter it into official contests.

[@quanghoangminh7626]

MAKE SURE IT IS STOCKFISH 8 64 bit (3400 ELO). not the older ver of stockfish

[@tommihommi1]

Quang Hoàng Minh stockfish 8 on linux using 64 threads and 1GB hash size.

[@InXLsisDeo]

But without opening and end tables, which is a shame. The author also criticizes the fixed pondering time of 1 mn instead of letting the programs manage time by themselves. Still, A0 shows strengths unknown to machines until now.

[@oscaro.172]

Damn, I can't even beat Stockfish level 5 on Lichess haha

[@airwann]

Oscar O. I struggle to beat the level 3 damn

[@oscaro.172]

Not gonna lie, I've lost to level 3 a couple of times before they nerfed it.

[@larseunic]

I can't beat lvl 4.

[@flyingpenandpaper6119]

Damn, I have to see what I can beat now Edit: ok level 3 is too hard

[@MrBolaextra]

C'mon, level 5 is easy. It makes mistakes. I'm not a strong player, but I beat it once in 19 moves. lichess.org/D2917a8mNQ3M

[@rahulrajan7982]

2:25 Looks like alpha zero is teaching stockfish to play like humans

[@billschannel1116]

Watching Alpha Zero, its seems like it's just deciding on how it wants the board to look, and then making it happen. I am pretty sure that Alpha Zero is not playing regular chess. I think instead its in the process of making a mosaic that we will only discover when the boards of every game are put next to each other. And then we will find out it's an animation as we cycle the games through their moves.

We are so dead.

[@DarkestValar]

I love your channel man, you have very good commentary and explanation, cheers from mexico.

[@ChadKakashi]

"Stockfish resigned the game..." it's like saying God has failed to create.

[@ulissemini5492]

FYI alpha does use brute force, just a different algorithm + the intuition from it's neural network.

[@yeeshusrivastava9651]

In this position Stockfish Resigned the game!

human: deep mind, what is your favorite dish? 1 nano second after. Deep mind: stock fish.

[@mrblacky6882]

My dream match was Ivanchuk vs Stockfish, now its Alpha Zero vs Ivanchuk!

[@mrblacky6882]

Obviously u dont know Ivanchuk very good. He is rated 5k ELO just no one knows, not even him.

[@HawkOfGP]

He wouldn't get mated that quickly, but the game might be resignable by the 20th move.

[@speedskis777]

That queen to h1 move was so weird haha

[@deepakr1907]

can u post all of deep mind games

[@Mastouril]

yes please

[@tommihommi1]

Deepak R all 100?

[@slay-bz3oz]

that'd be great. all 100 for the next 3 months no joke

[@Cscuile]

They're being awfully secretive with there games for some odd reason. (Same goes with the specs and actual learning time of Alpha Zero) So only 10 games viewable out of the 100...

[@dontshootthecat]

specs and learning time are in the paper they released, no?

[@mickcoram3579]

I hope that one day Agadmator describes me as a "...valued member of the chess community"

[@thegorn]

All the programmers did for AlphaZero was solve for "what's the idea here?", "what is coming?" and "captures, captures, captures"

[@thomast8555]

Ok but can it beat Ivanchuck ?

[@agadmator]

+Thomas T Not likely :)

[@alphonseblackwood2930]

Hey Agadmator, found your channel a week and a half ago. You're a stud, keep up the great work! I really liked this AI battle. #suggestion you should put different chess engines against eachother once every two weeks and give your analysis. Have a good day man!

[@Traumtheater0]

Hope AlphaZero refutes The Berlin Wall next ;-) THEN it will get my respect!

[@Traumtheater0]

I'm German, so I can call it the Berlin Wall if I want to ;-).

[@sausage4mash]

It ditched the French after a few hrs apparently but as you say loves the Berlin , bit annoying as I play the French wondering if it is now refuted

[@fleecemaster]

Depends on your level, to say something is refuted doesn't really make sense, as to play the way alphazero plays requires that level of mastery. If French defense works for you at your level, then it's the right play.

[@Aeroldoth3]

7:00 Black queen d8-d5 check. White doesn't want to lose queen or rook, so trades queens. This frees black from checkmate and leaves them a knight and two pawns up.

[@anispinner]

I wonder if deep mind knows how deep it is.

[@tyrellwreleck4226]

this guy explains plans that normal people don't anticipate. keep making videos :)

[@azharudeensalim]

Very satisfying

[@davedevosbaarle]

Google DeepMind made a similar AI to learn the game of go (the latest version AlphaGo Zero is only a few months old). It has transformed go theory quite a bit in the past 18 months. 18 months ago an earlier version of AlphaGo played its first public games against the #4 player of the world, which was still a human at that time and it was expected to take at least a decade before computers would beat top humans. The AI won that match 4-1 and the latest AI version is even 1400 Elo points better than that earlier version! In particular, opening theory has transformed a lot in the past 18 months, because of Google DeepMind's AI. Perhaps more than in the 100 years before. The same opening theory transformation could happen in chess. Side note: a striking similarity between AlphaGo Zero and AlphaZero Chess is that both seem to play quite "human"-like. Extremely interesting!

[@davedevosbaarle]

AlphaGo Zero's initial opening moves are not completely alien. The very first moves are quite normal in human play. But fairly soon after, it plays moves that a human top player would never have played up to a year ago (and which have become mainstream in human play since). But I wouldn't say that those new moves are the reason that AlphaGo Zero is so strong. It's its apparently flawless positional evaluation. And from that quality, it is able to play bold and original positional moves that human top players would not have dared to play before. I'm not much of a chess player, but it seems that AlphaZero is particularly good at positional play too (at least that is what I understand from some chess commentators). I can imagine that it favours some openings over others, because of subtle long term positional advantages that it is able to sort of "sense" without going deeply into tactical variations (almost like a human). In go this quality would be called a superior positional judgement.

[@peterdao7346]

It has already changed the legendary game of GO, which it was created to tackle, called originally Alpha GO Zero. But it is a general purpose algorithm that can be applied to any board game with complete information. It changed GO, a game much older than chess and it will be the most powerful chess program.

[@jcb3883]

Suren vs agadmator while both commentating. Lol Just be careful with suren's "let's go back" skill.

[@lordadamson]

hahahaha hello chess louverz

[@sandeepsaroha7593]

It will be countered by "what's the idea here" skill

[@TG-to5nf]

WOO! Another Deep Mind Video :) Nice

[@BramCohen]

In your one video where the dog isn't on the couch there's barking in the background

[@Xenon777channel]

Agadmator could you make a video on Alpha Zero's favourite variation against itself? They show in the PDF paper what it likes to play, which is the English with e5 i.e. Sicilian reversed and the Slav declined this line - 1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Nf3 a6 5. g3 dxc4 6. a4, Which is played in this game - Iljin, Timofey - Kamsky, Gata 1-0 D15 TCh-RUS Blitz 2016 1. d4 d5 2. c4 c6 3. Nc3 Nf6 4. Nf3 a6 5. g3 dxc4 6. a4 a5 7. e4 Bg4 8. Be3 e6 9. h3 Bxf3 10. Qxf3 Na6 11. Bxc4 Nb4 12. Qe2 Be7 13. O-O O-O 14. Rad1 Qc7 15. Bf4 Qb6 16. Be3 Qc7 17. Bf4 Qc8 18. g4 Re8 19. Bg3 Qd8 20. g5 Nd7 21. h4 Nb6 22. Bb3 Bd6 23. f4 c5 24. dxc5 Bxc5+ 25. Kg2 Qe7 26. h5 Nc6 27. Nb5 Rad8 28. g6 Kh8 29. gxf7 Rf8 30. Qg4 Rxd1 31. Rxd1 Qxf7 32. Bxe6 Qe7 33. f5 h6 34. Nc7 Qf6 35. Rd2 Ne5 36. Qf4 Nbc4 37. Bxc4 Nxc4 38. Re2 Bd6 39. Qg4 Bxg3 40. Qxg3 Ne5 41. Nd5 Qg5 42. Qxg5 hxg5 43. Nc7 Rc8 44. Ne6 g4 45. Kg3 Rc4 46. Nf4 Kh7 47. Ng6 Nf7 48. Kxg4 Rxa4 49. Kf4 Nd6 50. Ne5 Rb4 51. Nd3 Rd4 52. Ne5 a4 53. Ng4 b5 54. h6 b4 55. hxg7 Kxg7 56. Ke5 Rd3 57. f6+ Kg6 58. Rg2 Nf7+ 59. Ke6 Ng5+ 60. Ke7 Rd4 61. Ne5+ Kh6 62. Nf7+ Nxf7 63. Kxf7 Rxe4 64. Rh2+ Kg5 65. Kg7 Kf5 66. f7 Re7 67. Rh4 a3 68. bxa3 bxa3 69. Ra4 Ke6 70. Ra6+ Kd5 71. Rxa3 1-0

[@peeyushgoyal2239]

Just analysed this game using Stockfish....believe me stockfish suggests none of the given moves for itself

[@tomwohler6553]

Peeyush Goyal both engines had only 1 min to calculate for every move, maybee thats why Stockfisch chose a diffrent approach

[@tomyman]

do you have a 60 core machine? i think you don't understand how computer chess works

[@slay-bz3oz]

No, he's right. Stockfish was deprived of it's opening knowledge during these games, and was left with calculation only. The point of these games was to demonstrate how Alpha Zero teaches itself chess and calculates less, it knows what the good moves are like a human. Stockfish, calculating millions of variations, and alpha, "only" tens of thousands. I'm sure the games would be more drawish if they allow stockfish it's opening knowledge, but we'll have to see in the future.

[@slay-bz3oz]

When you analyzed the game, did your Stockfish suggest the moves and sacrifices Alpha played? And how did your Stockfish evaluate game.. what move lost the game according to your Stockfish?

[@stagna1959]

engines are idiots, they use only brute force.Carlsen said it, has to be true. , If you accelerate it million times, it is fast idiot, but still idiot. Alfa Zero actually is able to create something of its own , which is whole different league in machine chess. AZ didnt have opening book either, it created its own book in learning process. To complain of lack of cheat sheet is amusing. It is like some WGM would complain that she is not allowed to use tablebases in order to checkmate with knight and bishop. Such an event actually occured, WGM didnt know how to do it !

[@thephilosopherofculture4559]

The new element in chess that Alpha Zero has added is that the ability to move pieces takes rank over the traditional count of value of pieces.

[@curiousityfreak1783]

I am here from Russia and I can verify we just bought it in exchange for promise to not interfere with the next Trump elections ;)

[@walkers975]

FAKE NEWS

[@dandiaz19934]

Ouch, that one hurt! haha :(

[@islomkayumov4552]

Dan-The-Man that's what she said

[@jeremyhulbert3343]

No interference needed. Democrats will defeat themselves, as usual.

[@kdhavle]

You win the internet with that!

[@harshvardhansinha5267]

"Top tier Stockfish" Leela and Alpha Zero: Future is now, old man.

[@alankilgore1132]

I was reading a SF blog and the following information was presented... 1. DeepMind AlphaZero (DMAZ) was allowed to use a book, SF8 was not (unsubstantiated). 2. The 4 TPU's that DMAZ used are equivalent to 400+ CPU cores, so the DMAZ hardware was already 8x+ faster than the hardware that SF8 played on. 3. SF8 was only given 1Gb for hash use. This is consider far too little when 64 cores are involved. 4. Was SF8 able to use 6-man Syzygy endgame tables? (unknown) Sounds like the test was "fixed" to insure that DMAZ could win. In the world of chess engines, you can't compare apples/oranges if one side has the rotten fruit. The TCEC tourney forces all engines to run on the same hardware. What will happen if/when someone puts together a chess engine that can do move generation using a GPU? Will that be considered "fair"? Would it be allowed? This isn't like the 70's when all the top engines ran on different hardware platforms. They didn't ban Belle when they entered with custom chess hardware, but it isn't the same now. More and more people have access to incredible hardware, and TCEC, for example, makes that hardware available to all entrants. But where do we draw the line. When Deep Blue played Kasparov, they had some dedicated chess hardware. Without it, they probably would not have won. I know i'm arguing both sides, but you have to look at contests between engines through the eyes of what you want to accomplish. TCEC has evened the playing field; it is more a comparison of the software developers rather than who can provide the best hardware platform. For example, your engine will never be competitive now days unless you support multi-core, sorry, that is the way it is. Did DeepMind ask the SF team if they agreed to the conditions of the match? I'm guessing not, and if so, the hardware inequalities would have been addressed. I could be wrong...

[@Maharani1991]

+

[@miscelanea5351]

I believe that such imbalance can be argued with the fact that both engines are using very different approaches for the same given problem (winning a chess game). Think about it. Deep Mind is using a beautiful novelty: an algorithm that improves on its own.

[@Cscuile]

I completely agree with you! I hope others will spread the word. It seems a lot of people are misinformed.

[@roqsteady5290]

Probably from the alpha zero teams point of view, they just wanted to test their machine learning algorithms against the best program they had available at the time - and maybe they didn't have the optimal set up for running stockfish just hanging around in their office. What's impressive is not just that Alpha zero was able to beat a really strong program, but the fact that it did with much less reliance on brute force calculation and a seemingly impressive positional understanding. In the past, programs with sophisticated evaluation procedures have generally been inferior to programs that were optimised to look at more positions with relatively simple evaluation.

[@klausfpv3610]

Well it is possible that they didn't have the best platform to run SF on, but I can't believe they only had 1 gig of spare ram. This has to be intentional

[@JB_inks]

This is both very exciting and also absolutely terrifying at the same time

[@pokerandphilosophy8328]

The two engines only played each other for 10 seconds. That was enough time for them to play 744 trillion games.

[@ConnorMcCormick]

PN Houle source?

[@joppekoers3992]

dude, trust me

[@arunbabuktkmce]

Both are engines. The algorithm and PC power matters. Alpha played billions of games within 4 hrs and memorized the outcome. They made some functions to correlate unknown positions.

[@alejandrozunigaavila2398]

Alpha zero vs magnus!!!

[@infinitysalinity7981]

Cyanide vs baby :)

[@fleecemaster]

Still, I would like to see what Magnus's and other Grandmasters opinions of how alphago plays would be, I'd like to see their matches and what they think. It was fascinating to see what the masters of Go thought of AlphaGo

[@alejandrozunigaavila2398]

Fleecemaster magnus just loose the speed tournament there are many options

[@ЕвгенийСедашов]

I know that Stockfish likely was not as powerful as it could have been in this match due to hardware limitations and/or program version, but it is one beautiful game. And the Deep Mind's approach is just pure elegance since it essentially does not tell AI how to play the game but rather allows AI to figure how to play the game in a best way by itself.

[@ebwholesaler]

Alpha Zero "the Terminator" knows that by blocking that little baby black horsie, you condemn the black tower too. With one of his pieces, he blocks TWO of his opponent, almost for the whole midgame. Like playing a game 16 pieces AGAINST 14. - " GO BACK TO THE OCEAN, STOCKFISH ! "

[@filipkramaric6636]

Stockfish was running on 80 MILLION positions per second where as AlphaZero was running on 80 thousand per second. So Stockfish played 1000 times more games per second and still lost the match.

[@max-os6ex]

Yes, that's the power of deep learning. AlphaZero evaluates a lot less positions, but its evaluation of positions is much much better than the rules the humans coding Swordfish came up with.

[@WorldXHero]

Did you just assume Alpha Zero’s gender?

[@TanmayPatil37]

Alpha Zero is Apache Helicopter

[@avocado4662]

he's a boy.

[@6jackace]

Fuck off😂

[@SleepDaMouse-xd8dn]

It prefers pronouns that you couldn't even comprehend

[@kanck7909]

He did. so what? You mad? Lol

[@Trynottoblink]

Nice! You're on a roll, keep it up.

[@prafulherode7819]

I am getting scared of this artificial intelligence

[@kasparov937]

Praful Herode Imagine it hold the world hostage? Gets a hold of the nuclear codes.

[@InXLsisDeo]

Yes, there are good reasons to be so. Read the article "intelligence explosion" on Wikipedia. "If a superhuman intelligence were to be invented-either through the amplification of human intelligence or through artificial intelligence-it would bring to bear greater problem-solving and inventive skills than current humans are capable of. Such an AI is referred to as Seed AI[3][4] because if an AI were created with engineering capabilities that matched or surpassed those of its human creators, it would have the potential to autonomously improve its own software and hardware or design an even more capable machine. This more capable machine could then go on to design a machine of yet greater capability. These iterations of recursive self-improvement could accelerate, potentially allowing enormous qualitative change before any upper limits imposed by the laws of physics or theoretical computation set in. It is speculated that over many iterations, such an AI would far surpass human cognitive abilities." ... "AI researcher Hugo de Garis suggests that artificial intelligences may simply eliminate the human race for access to scarce resources,[15][27] and humans would be powerless to stop them.[28] Alternatively, AIs developed under evolutionary pressure to promote their own survival could outcompete humanity.[19] Bostrom (2002) discusses human extinction scenarios, and lists superintelligence as a possible cause: When we create the first superintelligent entity, we might make a mistake and give it goals that lead it to annihilate humankind, assuming its enormous intellectual advantage gives it the power to do so. For example, we could mistakenly elevate a subgoal to the status of a supergoal. We tell it to solve a mathematical problem, and it complies by turning all the matter in the solar system into a giant calculating device, in the process killing the person who asked the question. A significant problem is that unfriendly artificial intelligence is likely to be much easier to create than friendly AI. While both require large advances in recursive optimisation process design, friendly AI also requires the ability to make goal structures invariant under self-improvement (or the AI could transform itself into something unfriendly) and a goal structure that aligns with human values and does not automatically destroy the human race. An unfriendly AI, on the other hand, can optimize for an arbitrary goal structure, which does not need to be invariant under self-modification." Note that this isn't any different to how humans behave. Some of us are ready to kill for money, and obviously billionnaires like Trump or Putin never seem to be satisfied with enough money. But an IA would do so in a much more efficient way than we ever can.

[@andrewmims6898]

Please, more Deep Mind videos! Enjoyed them very much sir!

[@ri10q44]

Why doesn't two engines playing against each other always lead to the exact same game? I thought the point of an engine was to always play the best move. Wouldn't this mean that every game would be identical if played by 2 engines? Can anyone explain how this works?

[@curiousityfreak1783]

Ri10q There are often at least a couple of roughly equal moves, so it just chooses one of them randomly I guess.

[@evgiz0r]

there is randomness involved in order to avoid being exploited

[@kinjasgaliya1847]

Ri10q kzbin.info/www/bejne/oZWtn6Woj8qhnZI

[@fulltimereader1075]

There are different best moves depending on what strategic plan is in place.

[@LorenzoStroe]

Don't read any of the comments, they clearly don't understand. Computers cannot calculate all the possible moves, meaning they can actually play moves that are not the best compared to another engine that can ''see'' further. They can definitely calculate very deep but not infinitely deep. They also have to evaluate each position which is again more differences between engines.

[@mas3737]

So awesome! I wish they would play a thousand games, but I am sure that there are better uses for Deep Mind.

[@walkers975]

Stockfish 2400 seems more accurate

[@OneDerscoreOneder]

Walkers why

[@Kaizsu_]

OneDerscore One Because Alpha Zero made Stockfish look weak

[@AndersonNeo12]

Agad: "this is easily winning for white" [Stockfish plays Knight to F5] 8:06 Me: "gimme that pawn on B5!" [Stockfish - Knight to D6] Agad: "Congratulions, you are an excellent blunderer" ☺ ps.: It was really my first thought of instantly taking that juicy pawn and completly missing the fork.

[@ΑρηςΜακρης-π4τ]

Dont let the engines rule the human mind...

[@pkundrat]

It is too late for that sentiment :)

[@larseunic]

Waaaaaaay toi late

[@ΑρηςΜακρης-π4τ]

Peter Kundrat yes indeed...

[@ΑρηςΜακρης-π4τ]

Adrien Pompon right...

[@ΑρηςΜακρης-π4τ]

Grey Haven xDD

[@shadowdancerRFW]

Sacking the knight early on is the move which says it all. Stockfish is just a computer program, while AlphaZero is an entirely different entity. Razor sharp gameplay after losing pieces is something AlphaZero seems to do all the time, it plays for position, not for material gain, and that's the lesson for us to learn. In chess, position is more important than what pieces you have.

[@dockman3363]

Clickbaited again... I thought that the queen will be saced instead of the knight.

[@agadmator]

+Thunderous Void It says piece in the title, why would you think queen? And what do you mean AGAIN? :D

[@therealsachin]

hmm... I thought the title made it clear that it was a piece and not a queen.

[@dockman3363]

agadmator's Chess Channel first of all, isn't a queen a piece on the chess board? Secondly, I didn't mean clickbaited on you channel, I meant on youtube in general, and lastly, do you still play on chess.com? Because you're always online but always unavailable... Btw, I'm a huge fan of you videos, so please don't take in of this seriously 😒

[@Kuzune]

The Queen is a piece on the Chess board, however, not all pieces on the Chess board are Queens

[@dockman3363]

Kuzune I don't like oversmartness, so please stay out of this, besides, this doesn't change the fact that I got clickbaited...

[@shaurya0]

'Stockfish resigned'. Wow, that is very satisfying to say!

[@johngrey5806]

"Stockfish resigned the game." - I love that quote!

[@praqash]

In my opinion, it doesn't matter 4hrs or 2 days for a powerful processor which Alpha Zero has. It did all the calculation it had to do, in that 4hrs. Also, Alpha Zero didn't learned chess(as oppose to common notion) but it created a database of its own by generating millions of different combinations/possibility/line with results associated to each combinations/possibility/line. Which will later help it to pick the correct line based on the result.

[@abdullahalmuhit6921]

The defeating Magnus joke was pure gold :.)loved the video and it only made it better.

[@pcl5636]

Awesome game again! More of these please :)

[@elementsofphysicalreality]

Wow I’ve never seen so many good moves in a row. The tactics on the H file were also pretty neat.

[@SimplyChaotic98]

I really dig these computer vs computer games

[@MadaxeMunkeee]

I came back to this video 5 years later just to hear the original “stockfish resigned the game”

[@peters972]

Talk about “pin-to-win!”. Bobby fisher would be proud. Deep mind is really cashing in on that and seems to capitalize on both development of its pieces and/or suppression of development by the opponent, more so than just numerically based criteria. I guess that’s what makes these AI seem so human at times. However, I’m sure this is old news!

[@knirb9]

These are awesome! More of these!

[@joaolucasbraga4642]

It's just shocking the way Alphazero plays. We have a lot to learn from it. Maybe some defenses like the french are just losing for black lol but maybe not, who knows. It's amazing how it completely squeezed out stockfish with that incredible king manouver (e1-d2-e3) in Karpov's style. And also this astonishing knight sacrifice, completely paralizing the black's pieces, very Tal-like. This is all so awesome! All hail the mighty Alphazero!

[@PeterSedesse]

It understands positional value much better than stockfish which just grinds down material. The knight and rook were greatly devalued in this game, white knew it

[@gmsi7d371]

this is amazing. 4 hours of learning starting from scratch and then it beats the best chess engine ever. what the f*** !

[@eshneto]

Alpha 0 is teaching us a lot of chess. Keep up with these videos.

[@gluonpa6878]

Stockfish resigns the Game !!! Yeah, quite satisfying. Now I wanna hear "A0 resign the game" !

[@BarbeRace]

"Stockfish resigns" soooo satisfying!!!