Actually on tcec they're testing a Leela stockfish hybrid. One that goes NN during the middlegame and at a certain piece count, (for ENDGAMES) it switches to brute force for digging. I wonder if its gonna be better than either of the two.
It's not a leela-stockfish hybrid actually, it's a leela-leela hybrid (net 11248 and net 32072 for endgames). But a leela-stockfish hybrid would be interesting too for sure.
Why are they mixing two Leela IDs? Isn’t the newest ID generally the best one, or does that mean Leela used to be better at endgames or middlegames? Sorry I don’t know much about Leela.
@@neloangelo7297 AlphaZero wasn't built to run on standard operating systems. It plays chess in a different environment than you see here: it plays supercomputer vs. supercomputer games, whereas the games in this video were played on pc vs. pc. It's likely possible that Google could develop an open-source version of AlphaZero that could be run on consumer hardware if they were willing to devote the resources towards doing so, but such a version would be much less powerful than the real AlphaZero, since both the scale of the AlphaZero model and the Monte Carlo Tree Search (MCTS) method that it uses demand large amounts of processing power to be effective. Leela (or Leela Chess Zero), the engine shown in this video, is an open-source community-trained chess intelligence that is modeled after AlphaZero's architecture and designed to be run on consumer hardware. It's currently a work in progress, but it is created to be like AlphaZero but for your home computer. Hope that helps.
I did my own quick analysis on Lichess, letting Stockfish run to about depth 30 on each move (Stockfish's initial graph can't be trusted in games as sharp as this). It was dead even until Leela's Nxh4, which turns out is a losing move. However, that wasn't evident until depth 30, and required Stockfish to make 8 only moves in a row to maintain the winning advantage. After Nxh4, Leela actually played the best possible move for the next 15 moves according to Stockfish, but by then Stockfish's advantage was over 2 and the game couldn't be salvaged.
@@lmabacus404 thanks . When stockfish trapped the Leela's bishop in the corner: 25.) c6, why couldn't Leela just take c6? That seems to be the losing move imo.
@Alan G I really couldn’t tell you, but I let Stockfish analyze it and here’s what it found. TLDR Bxc6 is losing because after dxe5 too many pieces are unprotected, the c-file opens white has a winning attack. Black will have to give up more material. Qxc6 isn’t so bad but just simplifies the position with trades on c6 and Stockfish is just up a piece. Ba8 allows Leela to regain material with a fork, but the engine might be overvaluing material which is why it prefers that move. Stockfish thinks that after Ba8 the position is even until depth 25 when it starts to realize Stockfish is better. By depth 35 it’s up to +1.5. What’s interesting is that the second best move, Qxc6, gives a very similar evaluation by depth 35 despite initially being seen as way worse. I think part of it is Stockfish (and most engines) being bad at recognizing the long term harm of having a dead piece. Bxc6 is immediately losing (+5.7) after dxe5, attacking the bishop on f6 and with a discovery on the queen. Stockfish suggests Qc4 Rc3 and you’re losing the queen. Any other move and Stockfish suggests advancing pieces towards black’s king (ex. Nf4) Basically, by taking on c6 with the bishop, too many of your pieces - both bishops and the queen - are exposed. The c-file becomes half open and black’s King is in serious danger, and you have to give up material. Black will get mated pretty soon. I don’t know what’s going on with Qxc6. Stockfish says just trade pieces: Qxc6, Qxc6, Bxc6, Rc3, Bxf3, Rxf3. At the end of it all black wins a pawn, with two pawns for the knight and it’s roughly +2. I guess the position just simplifies and Leela is simply worse after getting nothing out of the Nxh4 sacrifice. I don’t know why Ba8 is better, since Leela is down more material in this like with a trapped bishop. I think it’s because as black you have e4 forking the knight and rook, but you’d still basically be down a piece since the bishop is trapped. I’m sure someone with more knowledge of chess could analyze that better as I’m only an amateur.
Amazing match. Amazing. I like the result. Still showing that Stockfish is king but at the same time, showing that Leela is definitely up there nearing Stockfish quickly. Letting Stockfish know that the days of crushing Leela are over. The rook vs 2 minor game though. Feel like that should have been a win but with that time pressure in such a hellish endgame position. But kudos to Leela and Stockfish.
@@badereric I mean, didn't Google's AlphaZero, a neural network AI, completely crush Stockfish? Both last year with Stockfish 9 and again recently with stockfish 10. Was controversial in 2017 because Stockfish wasn't given a book, but AlphaZero still crushed Stockfish recently even with an opening book recently. So, it doesn't seem farfetched that neural network AIs will start to dominate brute force AIs.
@@thorjelly AlphaZero didn't beat Stockfish, Google _claims_ that it did. You're falling into their advertising. As long as they refuse to conform to standard computer tournament rules and publish ways to actually reproduce the matches, no one should consider that to be true. I find it pretty amateur that Stockfish always had some handicap in the published games, definitely shows there is something to hide.
@@trevorrogers95 who doesn't idk how it came to be but Leela has this "everyone's little girl" vibe. And ppl referring to Leela as a she and her. Makes it cute. And she's AI not a regular engine so..
Who wouldn't be biased against the old school of chess engines. If I was old enough to follow chess for deep blue I would've cheered against Kasparov too. We will reach the heavens only by burning everything that we are.
The endgame reached around 1:09:00 is a fortress draw, but the fortress only works because White's g-pawn has moved. A similar fortress with g6-h5 pawns and dark-squared bishop holds even with unmoved White pawns, because the pawn break g4 will always result in a drawn KRP vs KBP endgame after Black plays hxg4. The problem with the h6-g5 fortress in the video is that when White plays the pawn break h4, Black is the one who is left with the edge pawn in the resulting KRP vs KBP endgame, but it is only a loss if White's pawn is still on g2. Example: 8/6kb/7p/4K3/8/8/6P1/R7 White wins with 1.Ra7+ Kg6 2.Rxh7 Kxh7 3.Kf6! reciprocal zugzwang 3...h5 4.Kg5! h4 5.Kxh4! Kh6 and White wins with the spare tempo 6.g3!
Please make a video trying to analyse the second game (assuming my short-term memory wasn't affected by overflow issues by watching this), with those sweet sacrifices that got balanced out like 10 moves after. Or you should probably ask google Assistant, if she can talk the moves over with Alpha zero
Your Commentary is Great : Instructive and there for, Entertaining. You should propose yourself to commentate the lives feeds of future tournaments I think all chess fans would appreciate you.
How far away are chess engines from realizing draws are the long term outcome in most positions? I mean as good as they are at chess they fail to see so many things
It's amazing how quickly ppl can think while playing chess at high levels! I watch Jerry narrating and playing out the possible outcomes and I can't even follow a simple pawn move :( Oh well, more study needed to get to that level I guess...
In game 5, is there some way to forcibly give back the exchange in the rook vs. bishop ending? I feel like if white can force the black king to recapture in a way that allows them to bring their king to f6 then white is clearly winning
Bro I really like your content you speak chess unlike some other channel I can relate to your chess lingua. You analysis while the play is on actuall makes sense. Owk about this game(1st) this time around leela didn't not consider this moves of sacrificing pieces i think is was because the development by stockfish was on time ! and flexible
@@dmcdouga07 No, it's not. It's just built using the same principles. Completely different training though. Alpha Zero was trained on Deep Mind's supercomputer, while Leela is trained on community hardware.
@@renx81 I think AZ would win too but I'm not sure about crushing Leela as long as they run on equivalent hardware: we never saw any results from AZ outside of the matches ran on 4 TPUs. TPUs are a pretty big upgrade in compute to the consumer grade graphic cards used by Leela.
@@toequantumspace he is right seeing those unbeatable machines capable of doing precise movements thousand or millions of calculations,even magnus struggle with his app and the app is inferior than stockfish 8 , simply play each other between humans seems pointless unless you pkay it like entertainment as any game but not trying to do your best i.e not years of it
About time Jerry! LOL I've been waiting what seemed like forever for a new vid from you. I look forward to seeing your videos. Thanks for all your work. I love your channel!! Je-rry !Je-rry! Je-rry!
@@dankwarmouse6248 Those people are speaking about the past decade plus of computer chess, about which a reasonable case could be made for strong but boring chess. The neural nets have improved the interest level, at least for me.
@@dankwarmouse6248 i mean there really is no traps.. non of them will fall for any traps... its not much agressive and both size playing positional dont really like these games
I don't understand why Lella blundered her castle at the end of that last game. Even I can see that you would lose the castle with that move so how did she miss it?
Is it me or with the engines help the game is going in the direction of total draw? Even creating a slight better position in the middle game is almost always impossible to capitalize on that in the end game. Maybe Carlsen-Caruana 12 draws were a statement of the game right now and the only thing that could get a loss is human error with thighter time controls. What do you guys think?
After seeing these new AlphaZero and Leela Chess Zero games, I'm not so sure. Draws, maybe, but this is extremely exciting chess. If this is the direction we're going in, sign me up.
Any chance of a deeper analyzis of these games? Preferably with the help of *both* engines analyzing each of the key moves as deep as your system will go. ;-)
1:03:57 "White doesn't wanna take the bishop white wants to go here, and then threaten this, wants to cut off the rook, plant the bishop on this square and maybe eye up this move. Meanwhile these pawns are all passed." *robot takeover confirmed*
Am I the only one who thinks this agreeing on a draw is quite impressive. I mean, you have to evaluate not only the position but also the quality of your opponent to do it, right?
I might be wrong, but I think there is a problem with the learning model for Leela. Like a lot of human players, she is awesome fast in the opening, genius in the middle and kind of confused in the end. That's consistent with having a lot of data for beginning moves and less for endings. It's something Josh Waitzkin noticed in players and suggested training should start with end games, then proceed to openings and middle games, since end games are rarer.
@@Amezylst That is just blatently false. The "no opening and tablebases" only applies in training; anywhere else people are free to use either or both. Leela is certainly using an opening book here. Please go and educate yourself.
@@jellykin7161 I'm curious about this. I can't find any information about Leela using an opening book anywhere. There was talk about forking the project and including things like that but I don't think it's been done. Thanks!
@@markbegenisich1535 Like any other engine, leela can use opening books in match games. It's only in training games where the "no opening books or table bases" applies. In this match, you can see leela (and stockfish) moving really fast in the opening. I have downloaded both engines and neither play that fast unless you give them an opening book.
to me this match once again proves that the value of the bishop and even the knight should in general be more like 3.5 and not 3... even in endgames those pieces do ok in comparison to rooks if perfect moves are played.
Actually blinked and missed the opening
Judging from the moves both players were cheating.
i feel like they might've used an engine
@@_emik hard to say... Never can be too sure
They are bots...
@@Royquah Good pick up man, these guys just don't understand the content
Roy Quah Wooooosh
''THIS IS SO BEYOND ME! OH MY GOD!!"
-Network, Chess - 2018
Me with 85% of the things Jerry says. LOLOL
Timestamp?
@@sporadicdrive5884 1:17:06
@@Kochira Thanks! You're golden
I dunno. I think he coulda drawn it.
Stockfish shows us how to gain 1000 points in a half dozen games.
you know you're watching something crazy when Jerry's commentary regresses to "Oh...My...God..." and giggles (=
"this is not a rating we can trust '1500'" hilarious hahaha
I laughed when he said that, this video is chess comedy gold
Stockfish Sandbagging lol
@@phen-themoogle7651 lol it's funny because I do that on lichess and someone once told me the same thing. Hahaha
@Crwth Ghiikilzic Yes it's around 3400 elo minimum.
1 second after he say that i Saw this comment, lol😂
Wow, Jerry. Your enthusiasm really added to the entertainment value of this vid.
Imaginær watching it without sound xD
Hearing Jerry happy makes me so happy. :)
Leela: [sips water]
Jerry: "interesting. I like where this is going. What are we witnessing here?!"
Love it hahaha
Both of them play in the style of BOTvinnik.
Niccccccccce
Botvinnik would be ashamed
It's good that we got to see some Botwinning.
They are just FABiano Caruana!
1:16:30 - 1:18:25
especially 1:16:50 was my highlight of this video.
That was intense!
Even though I rooted for LeelaChess, I share Jerry's excitement for the last game.
Actually on tcec they're testing a Leela stockfish hybrid. One that goes NN during the middlegame and at a certain piece count, (for ENDGAMES) it switches to brute force for digging. I wonder if its gonna be better than either of the two.
It's not a leela-stockfish hybrid actually, it's a leela-leela hybrid (net 11248 and net 32072 for endgames).
But a leela-stockfish hybrid would be interesting too for sure.
Why are they mixing two Leela IDs? Isn’t the newest ID generally the best one, or does that mean Leela used to be better at endgames or middlegames? Sorry I don’t know much about Leela.
Why leela this leela that? What happened to AlphaZero?
@@neloangelo7297 AlphaZero wasn't built to run on standard operating systems. It plays chess in a different environment than you see here: it plays supercomputer vs. supercomputer games, whereas the games in this video were played on pc vs. pc. It's likely possible that Google could develop an open-source version of AlphaZero that could be run on consumer hardware if they were willing to devote the resources towards doing so, but such a version would be much less powerful than the real AlphaZero, since both the scale of the AlphaZero model and the Monte Carlo Tree Search (MCTS) method that it uses demand large amounts of processing power to be effective.
Leela (or Leela Chess Zero), the engine shown in this video, is an open-source community-trained chess intelligence that is modeled after AlphaZero's architecture and designed to be run on consumer hardware. It's currently a work in progress, but it is created to be like AlphaZero but for your home computer. Hope that helps.
@@philipchalmers3572 Alpha Zero did not run on a super computer when it played SF. AZ *trained* on a super computer.
Can you PLEASE do a detailed analysis of the last game? I really need it...
I did my own quick analysis on Lichess, letting Stockfish run to about depth 30 on each move (Stockfish's initial graph can't be trusted in games as sharp as this). It was dead even until Leela's Nxh4, which turns out is a losing move. However, that wasn't evident until depth 30, and required Stockfish to make 8 only moves in a row to maintain the winning advantage.
After Nxh4, Leela actually played the best possible move for the next 15 moves according to Stockfish, but by then Stockfish's advantage was over 2 and the game couldn't be salvaged.
@@lmabacus404 thanks man!
@@lmabacus404 woh!! Thanks for that. Ever think about uploading you're findings? Would be an interesting watch
@@lmabacus404 thanks . When stockfish trapped the Leela's bishop in the corner: 25.) c6, why couldn't Leela just take c6? That seems to be the losing move imo.
@Alan G
I really couldn’t tell you, but I let Stockfish analyze it and here’s what it found.
TLDR Bxc6 is losing because after dxe5 too many pieces are unprotected, the c-file opens white has a winning attack. Black will have to give up more material. Qxc6 isn’t so bad but just simplifies the position with trades on c6 and Stockfish is just up a piece. Ba8 allows Leela to regain material with a fork, but the engine might be overvaluing material which is why it prefers that move.
Stockfish thinks that after Ba8 the position is even until depth 25 when it starts to realize Stockfish is better. By depth 35 it’s up to +1.5. What’s interesting is that the second best move, Qxc6, gives a very similar evaluation by depth 35 despite initially being seen as way worse. I think part of it is Stockfish (and most engines) being bad at recognizing the long term harm of having a dead piece.
Bxc6 is immediately losing (+5.7) after dxe5, attacking the bishop on f6 and with a discovery on the queen. Stockfish suggests Qc4 Rc3 and you’re losing the queen. Any other move and Stockfish suggests advancing pieces towards black’s king (ex. Nf4) Basically, by taking on c6 with the bishop, too many of your pieces - both bishops and the queen - are exposed. The c-file becomes half open and black’s King is in serious danger, and you have to give up material. Black will get mated pretty soon.
I don’t know what’s going on with Qxc6. Stockfish says just trade pieces: Qxc6, Qxc6, Bxc6, Rc3, Bxf3, Rxf3. At the end of it all black wins a pawn, with two pawns for the knight and it’s roughly +2. I guess the position just simplifies and Leela is simply worse after getting nothing out of the Nxh4 sacrifice.
I don’t know why Ba8 is better, since Leela is down more material in this like with a trapped bishop. I think it’s because as black you have e4 forking the knight and rook, but you’d still basically be down a piece since the bishop is trapped. I’m sure someone with more knowledge of chess could analyze that better as I’m only an amateur.
Love the enthusiasm...never laughed at a chess analysis until now!
Amazing match. Amazing. I like the result. Still showing that Stockfish is king but at the same time, showing that Leela is definitely up there nearing Stockfish quickly. Letting Stockfish know that the days of crushing Leela are over. The rook vs 2 minor game though. Feel like that should have been a win but with that time pressure in such a hellish endgame position. But kudos to Leela and Stockfish.
Spoilers much
Don't read the comments before watching? lol
Leela will not ever be stronger than stockfish
@@badereric I mean, didn't Google's AlphaZero, a neural network AI, completely crush Stockfish? Both last year with Stockfish 9 and again recently with stockfish 10. Was controversial in 2017 because Stockfish wasn't given a book, but AlphaZero still crushed Stockfish recently even with an opening book recently. So, it doesn't seem farfetched that neural network AIs will start to dominate brute force AIs.
@@thorjelly AlphaZero didn't beat Stockfish, Google _claims_ that it did. You're falling into their advertising. As long as they refuse to conform to standard computer tournament rules and publish ways to actually reproduce the matches, no one should consider that to be true. I find it pretty amateur that Stockfish always had some handicap in the published games, definitely shows there is something to hide.
AHAHHAHAHAH min:47:04 "these guys are playng like world champions" xD
Loved this, thanks Jerry! Your enthusiasm is contagious! 😀
In the last game that knight on f5 was just almighty. Fantastic stuff, great commentary, thanks for the upload!!!
absolutely absurd. Each game has Leela play some crazy sac which no-one can see coming. Thanks for uploading and your commentary!
47:00 oh that jab...
"Oh what am I doing?" *switches leela chess onto bottom* pretty obvious bias here jerry
He loves leela.
@@trevorrogers95 who doesn't idk how it came to be but Leela has this "everyone's little girl" vibe. And ppl referring to Leela as a she and her. Makes it cute. And she's AI not a regular engine so..
we don't get to choose who we love
Who wouldn't be biased against the old school of chess engines. If I was old enough to follow chess for deep blue I would've cheered against Kasparov too. We will reach the heavens only by burning everything that we are.
@@whong09 I was with you until that last sentence there.
The endgame reached around 1:09:00 is a fortress draw, but the fortress only works because White's g-pawn has moved. A similar fortress with g6-h5 pawns and dark-squared bishop holds even with unmoved White pawns, because the pawn break g4 will always result in a drawn KRP vs KBP endgame after Black plays hxg4. The problem with the h6-g5 fortress in the video is that when White plays the pawn break h4, Black is the one who is left with the edge pawn in the resulting KRP vs KBP endgame, but it is only a loss if White's pawn is still on g2. Example: 8/6kb/7p/4K3/8/8/6P1/R7 White wins with 1.Ra7+ Kg6 2.Rxh7 Kxh7 3.Kf6! reciprocal zugzwang 3...h5 4.Kg5! h4 5.Kxh4! Kh6 and White wins with the spare tempo 6.g3!
Great video man. Your personality brings the game alive for the viewers
Please make a video trying to analyse the second game (assuming my short-term memory wasn't affected by overflow issues by watching this), with those sweet sacrifices that got balanced out like 10 moves after.
Or you should probably ask google Assistant, if she can talk the moves over with Alpha zero
I enjoy you passion for the game...fun to watch/listen to
For a moment there i thought this mini match set was gonna be like the world championships.
At 59:30 couldn't white play Bxc4 then if dxc4 Nxc4 and I am attacking both Queen and bishop?
@Jacob Johnson Yeah and then if I play Bxd6 then b5 and it's winning back the piece. Thnx
What did he say at 1:12:27? Idk how to spell it and every google search I try just brings me to Forsyth-Edwards notation
Another video with an analysis of some of those moves which did not make sense would be nice. Like when a side seems to offer a minor piece.
Nobody:
Minecraft villager: 7:06
Lmaooooooo give me a reply lol
so underrated XD
Where is Lichess Arena 9 commentary? I need it
twitch
@@szhzs6121 Can't find it. It's not playing.
Ok now it's here:)
"What's the hesitation for Leela. Let's go!"
How brash, Jerry.
Splendid upload.
Dude I love your enthusiasm.
Your Commentary is Great : Instructive and there for, Entertaining. You should propose yourself to commentate the lives feeds of future tournaments I think all chess fans would appreciate you.
How far away are chess engines from realizing draws are the long term outcome in most positions? I mean as good as they are at chess they fail to see so many things
It's amazing how quickly ppl can think while playing chess at high levels! I watch Jerry narrating and playing out the possible outcomes and I can't even follow a simple pawn move :(
Oh well, more study needed to get to that level I guess...
Check out Hikaru Nakamura playing Puzzle Rush.
Looking forward to a deep analysis of that last game😁
last game was INSANE
Are the engines making their own moves from move 1?
In game 5, is there some way to forcibly give back the exchange in the rook vs. bishop ending? I feel like if white can force the black king to recapture in a way that allows them to bring their king to f6 then white is clearly winning
Bro I really like your content you speak chess unlike some other channel I can relate to your chess lingua.
You analysis while the play is on actuall makes sense.
Owk about this game(1st) this time around leela didn't not consider this moves of sacrificing pieces i think is was because the development by stockfish was on time ! and flexible
Has there been a match between Alpha Zero and Leela yet?
@@dmcdouga07 No, it's not. It's just built using the same principles. Completely different training though. Alpha Zero was trained on Deep Mind's supercomputer, while Leela is trained on community hardware.
Anyways, AZ would crush Leela at this stage.
@@renx81 A0 has only crushed SF8 as far as I know, and Leela is strong enough to fish SF10
Archer143 its all mostly depending on computing power, so it’s difficult to say atm
@@renx81 I think AZ would win too but I'm not sure about crushing Leela as long as they run on equivalent hardware: we never saw any results from AZ outside of the matches ran on 4 TPUs. TPUs are a pretty big upgrade in compute to the consumer grade graphic cards used by Leela.
Well, that's it. I ain't playing chess ever again
Why not? We can learn from the bots and lc0 plays very interesting attacking chess. Better donate some gpu time to improve leela
@@toequantumspace he is right seeing those unbeatable machines capable of doing precise movements thousand or millions of calculations,even magnus struggle with his app and the app is inferior than stockfish 8 , simply play each other between humans seems pointless unless you pkay it like entertainment as any game but not trying to do your best i.e not years of it
@@test-dn3pn Hikaru with Rybka and one pawn, was not able to win, against Stockfish, i think.
31:42 You counted that Pawn as 2 Pieces LOL 😂
The PGN s of this games are in Lichess? the game 2 its really cool
1 year later, I still doesn't get why Ba8 instead of BxC6
Thought it would be another Fabiano Carlson 12 draw match after those first two games
Been so long since I saw a Jerry review, gonna enjoy this at midnight :D
About time Jerry! LOL I've been waiting what seemed like forever for a new vid from you. I look forward to seeing your videos. Thanks for all your work. I love your channel!! Je-rry !Je-rry! Je-rry!
these games are absolute gold! its a shame to brush through them so fast. can we slow down a bit and get your thoughts on these positions next time?
How you manage to set stockfish 10 to lichess?
When are you going to upload the titled arena 9 commentary??
I'm just two games in, and it's already completely insane
Seriously, why do people say that computer chess is boring? These are some of the most amazing games I've ever seen.
@@dankwarmouse6248 Those people are speaking about the past decade plus of computer chess, about which a reasonable case could be made for strong but boring chess. The neural nets have improved the interest level, at least for me.
@@dankwarmouse6248 i mean there really is no traps.. non of them will fall for any traps... its not much agressive and both size playing positional dont really like these games
@@serxuegoldisprettyuselessi2683 Yeah but the depth in the moves is just beautiful.
28:47 interesting that stockfish didn't go for the King + Rook fork Ne5+ no?
Kd5 would counter attack whites rook.
watched the whole thing, really exiting stuff man
With Jerry's commentary I could watch paint dry :)
Hi Jerry
8:06
Yes it is!
I don't understand why Lella blundered her castle at the end of that last game. Even I can see that you would lose the castle with that move so how did she miss it?
Is it me or with the engines help the game is going in the direction of total draw? Even creating a slight better position in the middle game is almost always impossible to capitalize on that in the end game. Maybe Carlsen-Caruana 12 draws were a statement of the game right now and the only thing that could get a loss is human error with thighter time controls. What do you guys think?
After seeing these new AlphaZero and Leela Chess Zero games, I'm not so sure. Draws, maybe, but this is extremely exciting chess. If this is the direction we're going in, sign me up.
21:30 white to move
d5 check
bishop d4 skewer on the rooks
Why wouldn't that work?
Oh right, rook g3 check
30:18 "How do I know it's a he" :D you're funny :D
scan i download leela do my chess base, if so where can i get it???
1:13:35 - literally a quiet move
32:55 Leela G7-G2 or not, am i missing something ?
There's a fork
nvm, i see it now ther's a knight chek
Mini match? 1:26:28?
Well in classical chess one game can last up to 7 hours so six games in 90 minutes is very comfortably a mini-match.
You've seen that last ending so many times when talking about ways to describe infinity
8:10 yes Jerry.. "This chess stuff".. Right.great fun. 😄
Any chance of a deeper analyzis of these games?
Preferably with the help of *both* engines analyzing each of the key moves as deep as your system will go. ;-)
Both are Stockfisch. Helping each other to get Rating poinst ... 1000 in 6 Games.
And as Always : great commentary
How do they get them to play a different game each time?
opening database for the sake of variety
I can't say I got much from the games, but it was so fun to hear Jerry genuinely happy :D
Most entertaining commentary since the infamous scholars mate. :)
This was amazing to follow!
what is Leela Zero current rating ?
1:03:57 "White doesn't wanna take the bishop white wants to go here, and then threaten this, wants to cut off the rook, plant the bishop on this square and maybe eye up this move. Meanwhile these pawns are all passed." *robot takeover confirmed*
please upload ur commentary on the lichess arena 9
That last game Leela was like
"Leelo". Big balls mate. Too bad it didn't pay off that would've been incredible
Leeeellloooooo hhhhmmmmJeeehhnnnkiiiiinnsssss
I like when u talk about weekends :)
Loving it! thanks
No title arena on TH-cam?
Am I the only one who thinks this agreeing on a draw is quite impressive. I mean, you have to evaluate not only the position but also the quality of your opponent to do it, right?
For humans, yes. I think the computers do it based purely on the position.
I can just tell your favorite engine is Lc0.
ohhh, 1 and a half hour of greatness :D
Love waking up to this
I might be wrong, but I think there is a problem with the learning model for Leela. Like a lot of human players, she is awesome fast in the opening, genius in the middle and kind of confused in the end. That's consistent with having a lot of data for beginning moves and less for endings. It's something Josh Waitzkin noticed in players and suggested training should start with end games, then proceed to openings and middle games, since end games are rarer.
Being fast in the opening is due to an opening book, not because of leela's chess knowledge.
@@jellykin7161 Leela doesn't use an opening book or endgame tables, likely just fast from the countless openings she's experimented with in training!
@@Amezylst
That is just blatently false. The "no opening and tablebases" only applies in training; anywhere else people are free to use either or both. Leela is certainly using an opening book here.
Please go and educate yourself.
@@jellykin7161 I'm curious about this. I can't find any information about Leela using an opening book anywhere. There was talk about forking the project and including things like that but I don't think it's been done. Thanks!
@@markbegenisich1535
Like any other engine, leela can use opening books in match games. It's only in training games where the "no opening books or table bases" applies.
In this match, you can see leela (and stockfish) moving really fast in the opening. I have downloaded both engines and neither play that fast unless you give them an opening book.
Jerry is a lovely dinosaur
what is that supposed to mean??? 🤔🤔🤔
Protected Method He is a lovely green dinosaur
1:22:02 why can't white play queen to b7?
to me this match once again proves that the value of the bishop and even the knight should in general be more like 3.5 and not 3... even in endgames those pieces do ok in comparison to rooks if perfect moves are played.
Do it''lol Cool video with great commentary as usual. Thank you.
How to do that on lichess plss
The last game was pretty great to be honest
Jerry, you absolutely need to analyse these games in depth.
That 2nd game was a beauty.
This certainly feels like a world championship...
Wish someone would talk about me the way Jerry talks about chess.
Easy enough, just hang around for a thousand years and then get really interesting with the help of advanced computing.
fun to watch Thx
still waiting to see a monster that is playing like combination of both lleela and stockfish)))
The ending at 1:11:00 would be a win for White if the h3 pawn was on h2.
Wrong. I think you mean "if the g4 pawn was on g2."
"This is so beyond me... why am I even here?!"