Thank you so much for those 17 years, of interesting chess content production. I probably have watch 90% of production since subscribing 10 years ago. Keep up the good work.
The Leelas trained on the odds starting positions are just so interesting how they maneuver the opening. Makes me wonder how Leela would act if she was trained on the Chess960 format, and if that would cause some interesting changes in her playstyle.
There are games on youtube of leela vs stockfish with chess960. And they're pretty nuts. Fischer random is truly the best kind of chess. The world chess championship for it was infinitely more exciting and awesome than the classical championship.
yeah, 960 leela would be awesome to see. Fairy stockfish is just completely unopposed in 960 rn, because there's no Leela to fill the "good competitor" category. leela understands positions very differently to stockfish as well so it's always nice to see a clash of ideas.
@@copper4eva That's true, but that's the Leela model that was trained with normal chess, then applied to chess960. The idea is to instead let leela train against itself using chess960 random starting positions instead of the normal chess starting positions, which might make it more flexible (?). And @whenthingsfly4283 absolutely, stockfish is so dominant in the category and it'd be great to see some new ideas in the space.
For a while I was playing stockfish taking knight odds. I found that Stockfish played a lot more aggressively and turned the missing piece into "free squares" to enhance the remaining piece placement. I honestly felt like taking knight odds wasn't giving myself 3 points but rather 1.5 or even 1. I felt very embarrassed by this. I was an OTB 1800 and set Stockfish to approx play at that level.
I believe this version of Leelaknightodds is trained exploitatively against a model of human GMs, and makes certain moves that are inaccurate but complicating -- worse against engines, but better against their model of human GMs compared to the "accurate" move.
@@ChessNetwork Since it's a neural network, not an engine programmed by a human, it's likely its internal evaluation is too complicated to boil down to a single number like we're used to.
I don't like second guessing grandmasters but can someone explain why you would play c6 instead of just taking the bishop in c4 straight away? He ends up with this terrible backwards pawn on d6 and the bishop on c8 loses so much potential.
The d5 recommendation by stockfish shows how useful c6 was. You really need to get in d5 and blast open the center before you get squished on the kingside with g4-g5
Thank you for this game. Kind of mind blowing for me to see Leela at 88% accuracy with 20 centipawn losses despite 0-0-0 and Lenderman at 90% accuracy despite 23 centipawn losses and 0-0-1 blunder. How are these things calculated to make Lenderman more accurate? Arithmetic and logic apparently don't apply. Jerry, my mentor, please explain this seemingly illogical outcome. A long-time and happy subscriber.
Leela is missing 1 knight and to compensate, decides to paralyze 2 pieces: a knight and a rook, and even the bishop is almost paralyzed. Basically the black player was playing without those 3 pieces during the whole game.
I was puzzled why the GM castled King sighed to begin with. The lack of the computer King Knight indicated to me that it would be throwing pawns down the board on that side.
@arnoudh6203 The lack of the King's side Knight meant that there was one less defender on that side of the board for the computer. This, in turn, caused me to think that the pawns would be more valuable as part of a rushing attack; which is exactly what the computer did.
Thank you Jerry. Question - Knights are universally treated as being worth 3 points of material. In king knight odds games like this, would the missing knight still be worth the full 3 points? As in, is it giving black 3 points of advantage? Does quicker castling rights and a potentially more easily opened F file reduce the burden of playing down a knight somewhat?
You’re asking good questions. I just don’t know. I sometimes feel like Leela is using the absence of the knight in some positive way. For example, being able to castle a bit quicker and as a result activate a rook quicker. For what it’s worth, in Stockfish 17’s eyes the initial position without a white king knight reads -3.9.
@@schroederluck7984 Yep. That's because knights are very important in the opening. I suspect stockfish's evaluation would've been less if it were a bishop that was missing instead, because bishops aren't as important as knights in the openings.
The creator has some stats online, the b1 knight missing has Leela perform at 2818, missing the king knight she performed 2902, so you're correct. At least against human players that aren't insanely strong
@@cptnoremac i believe computer would beat magnus most of the time with this handicap. i think computer needs to exclude additional 1-2 pawns in order to make it even.
@@GEM4sta This bot, specially trained for knight odds, is noticeably stronger at the odds than the bot that Navara played. Right now Magnus would probably win a Rapid match with this bot, but the goal of the project is to reach the point where it can win a Rapid match at knight odds from any human, even Magnus, and I believe that the goal is attainable.
Wouldn't top players like Magnus or Caruana learn by playing these hyper strong engines with piece odds? Could be a way to improve your chess, learning from a stronger opponent.
Stick a fork in human players, computers play in a higher dimension! (That said, I wonder how much of the human play was hampered by raw intimidation?)
It wasn't early. Trying to find that sequence of seven only moves would be extremely difficult. You can see the big shift in the evaluation at the end and he likely saw the same problems Stockfish sees.
Again, don't like how the human played, really dim and unambitious play when you have a knight advantage, no sharp attacking line, not even a try, defensive, waiting play which, of course, results exactly like it should. If you have any advantage, go use it, no matter against a human or an engine.
last game I valued Lendermans play 1100, maybe that was too harsh, but my general criticism still stays. This time he played against a highly dynamic opponent, a6,b6,c6,d6,f6,g6 he was only missing h6 for the chinese wall, this way he would have at least completed the meme. Sorry this one was for sure 1100 and I dont care what stockfish or you Lenderman defenders out there say
By showing AI games Jerry destroys the game of chess. Thank you. Even Bobby Fisher said chess is stupid because nowadays they don't innovate but watch AI. I am not giving thumbs, let Jerry ask the AI for a thumb from now on.
@@Gru39 When they were handing out the kool-aid and tin-foil hats, you got the extra large didn't you? It's a tool, goofball, and you should probably learn more about chess (and know that Fischer's name isn't spelled "Fisher") before commenting such delusional and uninformed things.
@@arnoudh6203 I should relax my outrage because I just realized that I also play Civ 6 vs the AI instead of playing multiplayer games vs human opponents. There is a divide in the community and many players hate the AI because is cheating, but this is chess not Civ 6
Thank you so much for those 17 years, of interesting chess content production. I probably have watch 90% of production since subscribing 10 years ago. Keep up the good work.
Thank you for your decade of support. I appreciate it. 👍
The Leelas trained on the odds starting positions are just so interesting how they maneuver the opening. Makes me wonder how Leela would act if she was trained on the Chess960 format, and if that would cause some interesting changes in her playstyle.
There are games on youtube of leela vs stockfish with chess960. And they're pretty nuts.
Fischer random is truly the best kind of chess. The world chess championship for it was infinitely more exciting and awesome than the classical championship.
I want to see Leela trained to play Crazyhouse.
yeah, 960 leela would be awesome to see. Fairy stockfish is just completely unopposed in 960 rn, because there's no Leela to fill the "good competitor" category. leela understands positions very differently to stockfish as well so it's always nice to see a clash of ideas.
@@copper4eva That's true, but that's the Leela model that was trained with normal chess, then applied to chess960. The idea is to instead let leela train against itself using chess960 random starting positions instead of the normal chess starting positions, which might make it more flexible (?).
And @whenthingsfly4283 absolutely, stockfish is so dominant in the category and it'd be great to see some new ideas in the space.
@@copper4eva960 is great, leads to a lot of wild tactics and insane positions
Your talent to analyze chess games is not limited to human chess.
You are a real gift Jerry!
Thanks lumosity. 👍
Hey Jerry! Glad you're still making videos. I always come back to your channel 👍🏽
👍
Thank you for these LeelaKnightOdds-dev games and the commentaries/lessons, Jerry! They're fascinating and beautiful
Loving these Leela videos, Jerry! Thanks as always
Lendermen must’ve resigned because he suspected his opponent was cheating…
I really like those videos! Thank you Jerry for the great content!
For a while I was playing stockfish taking knight odds. I found that Stockfish played a lot more aggressively and turned the missing piece into "free squares" to enhance the remaining piece placement. I honestly felt like taking knight odds wasn't giving myself 3 points but rather 1.5 or even 1. I felt very embarrassed by this. I was an OTB 1800 and set Stockfish to approx play at that level.
There is no shame in losing to a robot
really interesting thanks for sharing looking forward to more
Good evening! I had the same subjective experience.
So this just proves engines are so op they can chuck in piece odds against a gm and win laughing
This was amazing. Thank you!
12:33
Stockfish throwing shade at Leela's accuracy lol.
Yeaah stockfish probably loses lol
That's hilarious. Zero blunders, mistakes, and inaccuracies but somehow only 88%. I guess they lost 12% by giving knights odds. 😆
@CesarGomez-kp5lm If you mean with Knight odds, you're a few corndogs short of a picnic
@@huk2617 i mean stockfish vs gm with knight odds of cousrse
I believe this version of Leelaknightodds is trained exploitatively against a model of human GMs, and makes certain moves that are inaccurate but complicating -- worse against engines, but better against their model of human GMs compared to the "accurate" move.
These make me want to see how Magnus would fair against Leela.
These games are awesome!
I don't think we have access but I'd be interested to see leela's own eval vs stockfish at moments
Agreed. Would be neat to see it.
@@ChessNetwork Since it's a neural network, not an engine programmed by a human, it's likely its internal evaluation is too complicated to boil down to a single number like we're used to.
love these jerry
Enjoy your take, thx
Love these ai games
6:59 Is like the game of Ding vs Magnus where Ding misses mate in 2 due Magnus giving up the queen on h7.
the fact this game was even almost the whole time according to stockfish is mindblowing
It wasn't even until the very last moves
Liked that one Jerry. Reminde me of course of Morphy. Thanks.
More AI videos please. This was so eye-opening 😮
thanks jerry
I don't like second guessing grandmasters but can someone explain why you would play c6 instead of just taking the bishop in c4 straight away? He ends up with this terrible backwards pawn on d6 and the bishop on c8 loses so much potential.
The d5 recommendation by stockfish shows how useful c6 was. You really need to get in d5 and blast open the center before you get squished on the kingside with g4-g5
Thank you for this game. Kind of mind blowing for me to see Leela at 88% accuracy with 20 centipawn losses despite 0-0-0 and Lenderman at 90% accuracy despite 23 centipawn losses and 0-0-1 blunder. How are these things calculated to make Lenderman more accurate? Arithmetic and logic apparently don't apply. Jerry, my mentor, please explain this seemingly illogical outcome. A long-time and happy subscriber.
LiChess counts the removal of the knight odds as a gross blunder! Hopefully they will fix that at some point.
88% accuracy for Leela but 90% for the GM
That's because LiChess considers the removal of the knight odds to be a gross blunder!!!
⭐️👍⭐️
Leela is missing 1 knight and to compensate, decides to paralyze 2 pieces: a knight and a rook, and even the bishop is almost paralyzed. Basically the black player was playing without those 3 pieces during the whole game.
I was puzzled why the GM castled King sighed to begin with. The lack of the computer King Knight indicated to me that it would be throwing pawns down the board on that side.
Mate, you are 300
Why would a lack of a knight mean you can push pawns more easily?
@arnoudh6203 The lack of the King's side Knight meant that there was one less defender on that side of the board for the computer. This, in turn, caused me to think that the pawns would be more valuable as part of a rushing attack; which is exactly what the computer did.
Ben Finegold said it's harder to spot a minor piece rather than a rook because your bishops and knights checkmate
He is wrong. LeelaRookOdds is much easier to beat or draw than LeelaKnightOdds, though still quite strong.
I believe Jerry would demolish Leela
Thank you Jerry. Question - Knights are universally treated as being worth 3 points of material. In king knight odds games like this, would the missing knight still be worth the full 3 points? As in, is it giving black 3 points of advantage? Does quicker castling rights and a potentially more easily opened F file reduce the burden of playing down a knight somewhat?
You’re asking good questions. I just don’t know. I sometimes feel like Leela is using the absence of the knight in some positive way. For example, being able to castle a bit quicker and as a result activate a rook quicker. For what it’s worth, in Stockfish 17’s eyes the initial position without a white king knight reads -3.9.
@@ChessNetwork Haha so it’s even worse than just randomly losing a knight. Thanks for the explanation Jerry. Very interesting dynamic for sure.
@@schroederluck7984 Yep. That's because knights are very important in the opening. I suspect stockfish's evaluation would've been less if it were a bishop that was missing instead, because bishops aren't as important as knights in the openings.
The creator has some stats online, the b1 knight missing has Leela perform at 2818, missing the king knight she performed 2902, so you're correct. At least against human players that aren't insanely strong
Also shows how robotic evaluation is. No edge for white at all during most of the game.
can magnus beat this computer with handicap?
Of course. Would he win every game out of 100? Probably not. Would he win most of them? Probably.
@@cptnoremac i believe computer would beat magnus most of the time with this handicap. i think computer needs to exclude additional 1-2 pawns in order to make it even.
He would probably score 8.5/10, maybe 9/10. David Navara who is significantly weaker scored 7/10 but he also spent time studying the bot.
@@GEM4sta This bot, specially trained for knight odds, is noticeably stronger at the odds than the bot that Navara played. Right now Magnus would probably win a Rapid match with this bot, but the goal of the project is to reach the point where it can win a Rapid match at knight odds from any human, even Magnus, and I believe that the goal is attainable.
Hi Jerry
👋
Wouldn't top players like Magnus or Caruana learn by playing these hyper strong engines with piece odds?
Could be a way to improve your chess, learning from a stronger opponent.
wen darkmode?
Hey Jerry, go browns.
Stick a fork in human players, computers play in a higher dimension!
(That said, I wonder how much of the human play was hampered by raw intimidation?)
Nice one Jerry. Disappointed with the ending from the GM. Early resignation should never happen on that level imho
It wasn't early. Trying to find that sequence of seven only moves would be extremely difficult. You can see the big shift in the evaluation at the end and he likely saw the same problems Stockfish sees.
Once it reaches +0 that's resign time against a bot. Looks like the lichess computer is saying 0.7, you're not coming back from that
@@GEM4sta So you can resign after the first move
i like your videos, but that thumbnail is cursed
a
please change the board color
This is the best in my case. I’m going to stick with this one.
Again, don't like how the human played, really dim and unambitious play when you have a knight advantage, no sharp attacking line, not even a try, defensive, waiting play which, of course, results exactly like it should. If you have any advantage, go use it, no matter against a human or an engine.
He's a grandmaster and you're probably not, so keep your 2 cents.
@@rawrxd4919 i really don't care about such comments, don't even find them original
@@RoyGazoff The fact that you replied proves that the opposite is true
First
Congratulations, that achievement should feel like a enormous jurel!
Jerry analyzes AI by using AI 😂. And he understands the soul of AI why. Jerry is the mediator spiritual guru the speaker of AI emotion and plans
last game I valued Lendermans play 1100, maybe that was too harsh, but my general criticism still stays. This time he played against a highly dynamic opponent, a6,b6,c6,d6,f6,g6 he was only missing h6 for the chinese wall, this way he would have at least completed the meme. Sorry this one was for sure 1100 and I dont care what stockfish or you Lenderman defenders out there say
Sounds like you're having a tough time against 1100s.
the eval stayed even though? how brazen of you
@@ethansmusic9898 I do ♥
@@jibbod8499 it did and thats not a surpise
By showing AI games Jerry destroys the game of chess. Thank you. Even Bobby Fisher said chess is stupid because nowadays they don't innovate but watch AI. I am not giving thumbs, let Jerry ask the AI for a thumb from now on.
Appeal to authority, appeal to nature, appeal to emotion, ad nominem, strawman
@@huk2617 You better wake up what I am saying and open your eyes. He analyzing AI with AI and presenting it as if human. Seriously messed up stuff
@@Gru39 When they were handing out the kool-aid and tin-foil hats, you got the extra large didn't you? It's a tool, goofball, and you should probably learn more about chess (and know that Fischer's name isn't spelled "Fisher") before commenting such delusional and uninformed things.
Wdym presenting it as if human, at what point does he say that@@Gru39
@@arnoudh6203 I should relax my outrage because I just realized that I also play Civ 6 vs the AI instead of playing multiplayer games vs human opponents. There is a divide in the community and many players hate the AI because is cheating, but this is chess not Civ 6