A little bit of Morphy in my life A little bit of Euwe by my side A little bit of Riga is all I need A little bit of Tarrasch is what I see A little bit of Steinitz in the sun A little bit of Magnus all night long A little bit of Judit, here I am A little bit of chess makes me your man Maia Number Five
The one thing that I really found unnerving with Maia is that it doesn't burn any time (I was playing the Maia9 bot). The moves are coming immediately which not only feels 'computerish' but you also cannot think on your opponents time as it is normal in human vs. human games. This somehow ruins the human-like experience that it was meant to provide.
Yeah, it would be cool to see an implementation where time management is also adapted to a human style. Lichess saves time usage for every move, so it should be no problem to train a model on that data as well.
Jeroen van den Berg i think it would be cool to be able to set the time he would take for each move manually or maybe set a setting which slowly increases the time he takes to "think" towards mid game
Not a very good normal comment, but when I’m drinking, I always put on your channel to help me sleep. And I don’t mean that in a bad way. I still follow everything you commentate and your personality is the best in all the chess channels. Always replay videos from your channel. You’re great at your job and it’s just a bonus points that your voice is the last one of the night I hear before I go to bed.
What I want is a version of Maia which learns to predict my moves as the website mentioned, and then, when it plays against me it uses what it knows about my biases against me. Playing against something like that would probably be very informative!
This is very cool although I'd like to see it mimic the time taken by a human as well. It's a bit disconcerting when it plays instantly, especially in a blitz format
Would be really cool if they could simulate momentum/confidence (being on fire) and it's opposite: tilting! Like if you play maia lots of times in a row, and it loses a winning position a couple of times, it would go on tilt
Kind of interesting that the bot trained on 1100 games plays at ~1550 strength. Maybe by learning common patterns in the moves that many 1100s make, Maia tends to pick up more on their decent ideas than their mistakes, as the good moves 1100s make probably have more in common by their adhering to fundamental principles than do their mistakes, which might be more off-the-wall, unevenly distributed, and not so well-learned, such that playing Maia1 is like playing a committee of 1100's, which collectively plays above the individual strengths of its members. edit: probably also the time factor, if it's always playing instantly you essentially lose half your time by not being able to think on your opponent's clock. Although when I played Maia in rapid it played at human speeds.
Those things, but also, was this trained on 1100s on Lichess? Because 1100 on Chesscom is closer to 1500 on lichess. In addition, I can say that on Chesscom, the difference between an 1100 and 1500 acutally isn't a ton, 1500 is less likely to drop random pieces especially in the opening but otherwise they are pretty much just as bad positionally.
@@GEM4sta Maia1 was trained on Lichess 1100 games, but that is true about the Lichess vs Chesscom ratings. I actually think 1585 vs 1100 is quite a significant rating difference on Lichess--it's 15th percentile vs 60th on the site, or barely above absolute beginner to a low club-level player.
well played jerry, as usual fantastic intresting, engaging and educational content. thanks for the chess and the laughs, have a nice christmas when it comes
Imagine if this at some point could reference YOUR games, coach and teach you what YOUR common mistakes are, what patterns you often miss, then also create a training program that gradually improves your skills. And to boot generates some motivational banter and a human readable/visualization as to what you are doing wrong and why..
Honestly this sounds like it could totally happen within the year. With the way technology is progressing in general and how impressive this chess engine is, we could totally get something like your example very soon
Great video! I lost it at “is that a human move, lose a game and resign?” Jerry, you should totally capitalize on the recent chess boom by posting more videos! A lot of new players can surely get a lot out of your very pedagogic style of explaining chess. You’re the OG chess TH-camr! I remember getting into chess around the Carlsen vs Anand 2 match and if I’m not mistaken, only you and Mato Yelic were covering those games on TH-cam back then. I would also love to see you in a broadcast format. You would do a much better job at that than those new guys on chess24. Collaborations would also be cool to see. Maybe with Botez, Bartholomew, Yasser or Jennifer Shahade?
What you said about how the stockfish-levels work isn't really true. That was how it worked for old chess-computers from 30 years ago. Stockfish and the likes have way too good evaluation- and pruning-algorithms for measure like that having the intended effect. Also this way just led to very inhuman mistakes due to the horizon-effect. So there's other ways to get down to human-level other than using NNs that are trained on human-play. The way these super-engines are limited usually is by having an average below ideal-evaluation-goal for their moves, which is determined by analyzing the average accuracy of human games of the targetted ELO. So it still calculates the best moves but only picks them, when this move is closest to the evaluation goal. For example if the goal is to be 0.2 pawn-values worse than ideal play, then the engine will pick moves that were evaluated like that. The main reason why this feels not really like a human of that elo is mostly because humans mistakes don't average over a game but go down in endgames. So when you get to the endgame against a downscaled stockfish, it starts playing really weird and making very unhuman mistakes. I think this could be fixed relatively easily, if you multiply the average evaluation-goal with number of remaining pieces/pieces from the start. So increase the initial evaluation-threshold to compensate but multiply it by that remaining-pieces-factor. So at 5 pieces it should only play 0.3 * 5/16 pawn-values worse per move on average instead of 0.2. Of course that would still not be as humanlike as a NN trained on human games but it would make the distribution of errors on lower levels more believable.
I think someone did some sort of research with bots to mimic their play and opening lines. I don't remember how many games were played. I think they did it at 1975 style and Fischer bot won. I think they did it again in 1978 style and Karpov won the rematch.
The moves may be at a 1900 level, but the engine is cheating you a little bit. It is moving instantly and stealing your time to think on your opponents turn. If would be more fair (more like a real 3 minutes each blitz game) to give yourself an extra 2 minutes to make up for this fact.
If Maia can perfectly replicate players another cool application would be to bring past chess heroes back to life, like Fischer, Spassky, Capablanca etc.
It can never _perfectly_ replicate their play, but you're right that it could be used to simulate past players with a higher degree of accuracy than ever has been accomplished before. But, yeah, I had the same thought as you. I suppose the accuracy will be higher with each game a player has in the database.
22:15 that move is actually considered innacurate and the best response is e6! White can force this line : f4 be7 8. qf3 or qd3 and black seems to have equalized
This is such a great idea, especially as it improves over time. It would make playing against bots so much more fun. Currently "easy" bots just feel like you're playing magnus carlsen, except for when they randomly give you a piece.
The only thing I don't like about Maia is that he's flying around. He should do some minimum 2 seconds per move to give us humans some time to think...
ChessNetwork Bingo (Ultimate Edition) - Hi everyone, it's Jerry - pop quiz \ - feel free to pause the video - with a punch - piece is 'sad' - A sad move to make, but what else is there? - piece has 'no future' - You wouldn't believe how fast this pawn can run - chop chop - this is very instructive - D E V E L O P Y O U R P I E C E S tall pawn / Tall pawn syndrome Biting on a rock That's not a 3 point piece anymore, more like a rook or even a 7 point piece That's very committal Light squares! Dark squares! Give him a boot! Ask the bishop a question "OH NO! I'm mated! Oh noooo! I got scholar'd!" - calling a piece an animal/monster - queen/rook goes into Pac-Man mode - calling a piece a dummy - "The queen goes pawn hunting" - as always, I hope you got something out of it, take care, bye The (...) will have to find a new home. Look how cozy white's/black's king is. The king walks. What is this piece NOT hitting? This (...) is a goner. This (...) is dead. The (...) is frozen. This (...) now has a responsibility. White/black says "NO". White/black says "NO WAY". True. But look at this structure. Space invaders! Queen trade? That's the home stretch. Chop-a-roni. I don't want to give up the (...) square. Fantastic post. Opposite-color bishops. This is a won king and pawn endgame. Watch out for back-rank mates! No stalemates! Where's the mate?! Is this mate?! Fancy tactics. This is now a passer. The pawn bolts. “Star move” “We’re only x moves out” “If we compare the pieces/kings” (Anything about the king’s “home”) “How do you meet this threat? You don’t.” “Some repositioning/shuffling back and forth” “This is just one variation/way this game could end” “There’s great symmetry... we look for imbalances” Credits: Slight Lokii, Phil Buttle, Chiotare, Robert Walter, carter, redorange, Menno Maman, Matheus A. Ferreira, None None, Juan Pablo Flores, Hydra.
33:45 Blind spot but maybe because of the backward Knight move syndrome, specifically there. So it could then be caught ahead of time, recognizing that common mistake. Great games and examples for that resource! Thx. Ps. I don't think Maia will dethrone the King, but just give us mortals a chance at mirroring him -- and making something great of our own play -- or mirror other greats too! He will continue to improve too I think, and is maybe the future goat, as they say -- the best ever! Thx again!
I love the content Jerry! This is so cool! I feel like next time I’d like to see you play without the urgent time control so you have time to think and talk out your moves though. Your analysis and critical thinking always amazes me.
I also want to see Maia chess try and flag opponents when in losing positions. Also it should play troll like openning moves when facing opponents whose elo is significantly lower. This is what Carlsen and high level players do on Lichess.
I recommend playing against lichess BOT "Boris Trapsky" - he plays well like a 2100 but creates unsound traps for an oponent :). The opening phase he plays is from a set of most played moves at lichess in that position :)
actually if people really wanted to cheat they can still play suboptimal human like plays with stockfish suggestions. Kind of a waste of time though, I do think as a result more tournaments should happen in person between a chess board.
@@primekrunkergamer188 it's hard to believe chess cheats enjoy being nothing more than a surrogate for a chess engine but they are out there. Guess it helps them feel better about themselves for being absolutely trash at the game
Well, this is probably not so human, in a bullet game when either sides time nears zero the bot is speeding up like hell and still finding good or only moves.
Maia learned from 12 million human games played online. And out of those 12 million games 10 million of them were cheating with an engine. So Maia is the same as Leelazero and Alphazero at the end of the day.
Maybe the moves are more human, but the way it plays, spending no time at all, is so inhuman and infuriating. Nobody will ever blitz out moves like that in 3+2, that's one of the main reasons I don't like bots. They should include some kind of waiting time after the calculation is over.
You are right. I prefer to play without time against bots. I am pretty sure they could simulate the time easily too, but you would have to get a big DB with the times recorded, and I can't think of any.
@@harleykf1 Wow. That would be great news. When I download games with the interface it doesn't show the times, but if you say they appear in the analysis, they probably have it, but don't send it when you ask for the pgn. That would be an amazing source of info for the Maia team, if they wanted to go in that direction.
@@alfa-psi dude he's saying it should not go "public", of course it is a great scientific achievement and should be used to improve our understanding of the game. He's just conveying the concern about online cheating which is a serious problem.
For now it's only 1750 blitz strength as a style matching alg like this is incapable of searching more than one move deep (since it has no eval function)
I feel like you were a bit too intimated by the fact that you were playing an engine, missing that its distinctive feature is that it blunders a lot (at least these versions, that were trained on blitz games). For example, at 25:21, you can win an exchange with Nxd5, at 26:28, you can win immediately with Rf1+ and Ra8+, and at 32:30, Nf6+ also ends the game immediately.
@5:40 I never understand why this pawn isn't hanging can someone please explain to me, I see grandmasters never take it but I don't understand why. I know the dragon is behind the f6 knight but I still don't see any discoveries because the queen is not on d2?????
Do you mean the e5 pawn by black? If you mean that, it is because after dxe5 dxe5 Nxe5 Black can play Nxe4! And now the material is equal and both knights are attacked, blacks and whites
The last game was pretty crazy! However, I almost jumped out of the chair when you didn't take on f6 immediately winning at 32:38 :P I guess great players also miss easy moves sometimes
Maia was never designed to play chess and chess is not its destiny. I was so intrigued to read their website. Looking forward to seeing all the different ways it can help. As it states on their web site: "Maia is an ongoing research project using chess as a case study for how to design better human-AI interactions."
You’re statement is misleading. It was directly designed with chess in mind, but they’re hoping to use the insights from this project for non-chess related activities.
@@Mecinimi Once you learn how to read, take as long as you need to read their website. Then keep re-reading it until you understand. If you still fail, maybe go and play in the sand pit.
I made a blunder against maia1 (it was rated 1600+ when I played) that would have been a sure loss against Stockfish at that level. The funny thing is I made a comeback and actually got to a winning position up 2 pawns but then blew it with letting it nail me with a perpetual check/draw :( Very human like game though! (compared to Stockfish) EDIT: I'm one of those mostly play casual against the computer people........but I'd estimate I'm about 1500~1600. Best win was 1760ish human and level 6 (2300) Stockfish, but that SF lv6 was sooo mentally draining to beat for me lol.
@ChessNetwork i recommend playing against lichess BOT "Boris Trapsky" - he plays well like a 2100 but creates unsound traps for an oponent :). The opening phase he plays from a set of most played moves at lichess in that position :)
I’m only rated around 1600 and I have two wins against Maia9. Both times Maia made blunders regarding discovered attacks. I feel like a human would evaluate these positions a little better. Maia plays too fast imo
I have to say that something like this could the end of honest online chess. Because if this algorithm could be used or mimicked and perfected, then a person could use the maia engine to cheat and then the online chess sites might not be able to detect it. Its a great technology but its unfortunate if it ends up just making it easier for people to cheat.
Hmmm that's interesting, but although it's human-like, you wouldn't make 2000 moves with only 1600 account.. You'll eventually blunder, so maybe harder than stock fish or Leela to detect but it's probably obvious. It's like letting magnus carlsen play with your personal account, although it's human-like, you definitely cheat with magnus beside you... When you, let's say, are only 1800.
@@diskyariajetmiko They have versions from 1100 to 1900 though. It's very plausible to use a lower level version of the engine and be fairly undetected
@@diskyariajetmiko well, the point is to make it human-like, and people think on the opponent move usually, you don't have time to think on your opponent turn if they play instantly. It's been modified anyways.
Maia: *loses full queen*
"Your opponent has left the game. You can claim victory in 30 seconds."
I honestly forgot how good you are jerry! Honestly you're the person who got me really into chess, so thanks!
Me too. He was there first chess personality that I followed
I think it was same here.
same here!
What’s his elo?
His Tal game where he takes his opponent into a deep dark forest is the video which got me into chess
A little bit of Morphy in my life
A little bit of Euwe by my side
A little bit of Riga is all I need
A little bit of Tarrasch is what I see
A little bit of Steinitz in the sun
A little bit of Magnus all night long
A little bit of Judit, here I am
A little bit of chess makes me your man
Maia Number Five
Why mention judit alongside such legends? She was never even in the top 5
@@KARTIKEYA007 I do it all for the music
@@KARTIKEYA007 The Judit part should've been, "a little bit of Capa, here I am". lol
The one thing that I really found unnerving with Maia is that it doesn't burn any time (I was playing the Maia9 bot). The moves are coming immediately which not only feels 'computerish' but you also cannot think on your opponents time as it is normal in human vs. human games. This somehow ruins the human-like experience that it was meant to provide.
Yeah, it would be cool to see an implementation where time management is also adapted to a human style. Lichess saves time usage for every move, so it should be no problem to train a model on that data as well.
Jeroen van den Berg i think it would be cool to be able to set the time he would take for each move manually or maybe set a setting which slowly increases the time he takes to "think" towards mid game
En Lucas Chess hay una opción para que juegue imitando tiempos humanos
Jerry: "lose a game and quit?!, is that the human move right there?" LOL so true
Not a very good normal comment, but when I’m drinking, I always put on your channel to help me sleep. And I don’t mean that in a bad way. I still follow everything you commentate and your personality is the best in all the chess channels. Always replay videos from your channel. You’re great at your job and it’s just a bonus points that your voice is the last one of the night I hear before I go to bed.
maia quitting after losing was the most human move in this video.
MAIA, I'm sorry if you learned from any of my games...
What I want is a version of Maia which learns to predict my moves as the website mentioned, and then, when it plays against me it uses what it knows about my biases against me. Playing against something like that would probably be very informative!
Anton squared you
at 32:29 was a beautiful knight fork to check and win the queen! I know it was high time pressure but just wanted to mention it :-)
This is very cool although I'd like to see it mimic the time taken by a human as well. It's a bit disconcerting when it plays instantly, especially in a blitz format
I just don't get why they didn't call them
Maia Hee
Maia Hoo
Maia Ha
and
Maia Haha
Maia Hoo would be the most used
Was looking for this comment, glad I found
maiasaurus
Would be really cool if they could simulate momentum/confidence (being on fire) and it's opposite: tilting!
Like if you play maia lots of times in a row, and it loses a winning position a couple of times, it would go on tilt
27:48 Maia becoming very human like, even learning to rage quit!
Kind of interesting that the bot trained on 1100 games plays at ~1550 strength. Maybe by learning common patterns in the moves that many 1100s make, Maia tends to pick up more on their decent ideas than their mistakes, as the good moves 1100s make probably have more in common by their adhering to fundamental principles than do their mistakes, which might be more off-the-wall, unevenly distributed, and not so well-learned, such that playing Maia1 is like playing a committee of 1100's, which collectively plays above the individual strengths of its members.
edit: probably also the time factor, if it's always playing instantly you essentially lose half your time by not being able to think on your opponent's clock. Although when I played Maia in rapid it played at human speeds.
Those things, but also, was this trained on 1100s on Lichess? Because 1100 on Chesscom is closer to 1500 on lichess. In addition, I can say that on Chesscom, the difference between an 1100 and 1500 acutally isn't a ton, 1500 is less likely to drop random pieces especially in the opening but otherwise they are pretty much just as bad positionally.
@@GEM4sta Maia1 was trained on Lichess 1100 games, but that is true about the Lichess vs Chesscom ratings. I actually think 1585 vs 1100 is quite a significant rating difference on Lichess--it's 15th percentile vs 60th on the site, or barely above absolute beginner to a low club-level player.
28:07 "Lose a game and you quit. Is that the human move" LOL
I'll believe this engine is human-like if it starts pushing pawns recklessly after blundering mate in 1 in the last game
hehexd
well played jerry, as usual fantastic intresting, engaging and educational content. thanks for the chess and the laughs, have a nice christmas when it comes
Thank you Kev and likewise. 👍🏼
Your voice really is perfect for chess commentary. Great games!
Imagine if this at some point could reference YOUR games, coach and teach you what YOUR common mistakes are, what patterns you often miss, then also create a training program that gradually improves your skills. And to boot generates some motivational banter and a human readable/visualization as to what you are doing wrong and why..
Honestly this sounds like it could totally happen within the year. With the way technology is progressing in general and how impressive this chess engine is, we could totally get something like your example very soon
@@justdakotamusic I'm really hoping for it. I feel like I'm in desperate need of something like that.
Chessnut Evo tiene eso!
Great video! I lost it at “is that a human move, lose a game and resign?” Jerry, you should totally capitalize on the recent chess boom by posting more videos! A lot of new players can surely get a lot out of your very pedagogic style of explaining chess. You’re the OG chess TH-camr!
I remember getting into chess around the Carlsen vs Anand 2 match and if I’m not mistaken, only you and Mato Yelic were covering those games on TH-cam back then. I would also love to see you in a broadcast format. You would do a much better job at that than those new guys on chess24. Collaborations would also be cool to see. Maybe with Botez, Bartholomew, Yasser or Jennifer Shahade?
What you said about how the stockfish-levels work isn't really true. That was how it worked for old chess-computers from 30 years ago. Stockfish and the likes have way too good evaluation- and pruning-algorithms for measure like that having the intended effect. Also this way just led to very inhuman mistakes due to the horizon-effect.
So there's other ways to get down to human-level other than using NNs that are trained on human-play.
The way these super-engines are limited usually is by having an average below ideal-evaluation-goal for their moves, which is determined by analyzing the average accuracy of human games of the targetted ELO. So it still calculates the best moves but only picks them, when this move is closest to the evaluation goal.
For example if the goal is to be 0.2 pawn-values worse than ideal play, then the engine will pick moves that were evaluated like that. The main reason why this feels not really like a human of that elo is mostly because humans mistakes don't average over a game but go down in endgames. So when you get to the endgame against a downscaled stockfish, it starts playing really weird and making very unhuman mistakes. I think this could be fixed relatively easily, if you multiply the average evaluation-goal with number of remaining pieces/pieces from the start. So increase the initial evaluation-threshold to compensate but multiply it by that remaining-pieces-factor. So at 5 pieces it should only play 0.3 * 5/16 pawn-values worse per move on average instead of 0.2.
Of course that would still not be as humanlike as a NN trained on human games but it would make the distribution of errors on lower levels more believable.
I think to be more like a human Maia should spend more time for move, playing always so fast is a huge advantage. Love Jerry’s videos.
Human: Learns computer moves
Computer: Learns human moves
What is this? Where are we going to?
We may finally have a Fischer-Maia vs Karpov-Maia match.
I think someone did some sort of research with bots to mimic their play and opening lines. I don't remember how many games were played. I think they did it at 1975 style and Fischer bot won. I think they did it again in 1978 style and Karpov won the rematch.
YaaaaY finally a new machine video from the best chess channel
32.30 you had nf6+, a little forkeroni winning the queen. I can’t believe you didn’t see that with a whole 4 seconds on the clock. SMH.
Shouldn't Maia Chess just let themselves flag after dropping a piece in a significant portion of their games? Seems pretty unrealistic otherwise.
Great video as always Jerry, thanks for the upload. Shame you had no time at the end! 37. Qxh3 Qxh3 38. Ng5+ forkaroni!
Up a pawn with a strong middle game, loses the endgame.
1900 player: My time has come
The moves may be at a 1900 level, but the engine is cheating you a little bit. It is moving instantly and stealing your time to think on your opponents turn. If would be more fair (more like a real 3 minutes each blitz game) to give yourself an extra 2 minutes to make up for this fact.
Maybe play with no time control. That way you can always play at your own pace and don't need to wait for the engine
If Maia can perfectly replicate players another cool application would be to bring past chess heroes back to life, like Fischer, Spassky, Capablanca etc.
It can never _perfectly_ replicate their play, but you're right that it could be used to simulate past players with a higher degree of accuracy than ever has been accomplished before. But, yeah, I had the same thought as you. I suppose the accuracy will be higher with each game a player has in the database.
I’ve missed some great ChessNetwork content. A great thing to wake up to :) thanks Jerry.
22:15 that move is actually considered innacurate and the best response is e6! White can force this line : f4 be7 8. qf3 or qd3 and black seems to have equalized
equal in a sicilian game that is^^
This is such a great idea, especially as it improves over time. It would make playing against bots so much more fun. Currently "easy" bots just feel like you're playing magnus carlsen, except for when they randomly give you a piece.
The only thing I don't like about Maia is that he's flying around. He should do some minimum 2 seconds per move to give us humans some time to think...
Forever great fan of Jerry. And Jerry's voice always chills me no matter what mood i am in. And once again waiting for your face reveal.
H2!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
I love how he plays with his food and then he gets to 20 seconds he like, “Let’s finish this.”
ChessNetwork Bingo (Ultimate Edition)
- Hi everyone, it's Jerry
- pop quiz
\
- feel free to pause the video
- with a punch
- piece is 'sad'
- A sad move to make, but what else is there?
- piece has 'no future'
- You wouldn't believe how fast this pawn can run
- chop chop
- this is very instructive
- D E V E L O P Y O U R P I E C E S
tall pawn
/ Tall pawn syndrome
Biting on a rock
That's not a 3 point piece anymore, more like a rook or even a 7 point piece
That's very committal
Light squares!
Dark squares!
Give him a boot!
Ask the bishop a question
"OH NO! I'm mated! Oh noooo! I got scholar'd!"
- calling a piece an animal/monster
- queen/rook goes into Pac-Man mode
- calling a piece a dummy
- "The queen goes pawn hunting"
- as always, I hope you got something out of it, take care, bye
The (...) will have to find a new home.
Look how cozy white's/black's king is.
The king walks.
What is this piece NOT hitting?
This (...) is a goner.
This (...) is dead.
The (...) is frozen.
This (...) now has a responsibility.
White/black says "NO".
White/black says "NO WAY".
True. But look at this structure.
Space invaders!
Queen trade?
That's the home stretch.
Chop-a-roni.
I don't want to give up the (...) square.
Fantastic post.
Opposite-color bishops.
This is a won king and pawn endgame.
Watch out for back-rank mates!
No stalemates!
Where's the mate?!
Is this mate?!
Fancy tactics.
This is now a passer.
The pawn bolts.
“Star move”
“We’re only x moves out”
“If we compare the pieces/kings”
(Anything about the king’s “home”)
“How do you meet this threat? You don’t.”
“Some repositioning/shuffling back and forth”
“This is just one variation/way this game could end”
“There’s great symmetry... we look for imbalances”
Credits: Slight Lokii, Phil Buttle, Chiotare, Robert Walter, carter, redorange, Menno Maman, Matheus A. Ferreira, None None, Juan Pablo Flores, Hydra.
Hahahahah
read all of these in his voice
Well said. ☺️
@@ChessNetwork Perhaps you could give us a video, with all your 'best hits''? :)
"ooh careful"
33:45 Blind spot but maybe because of the backward Knight move syndrome, specifically there. So it could then be caught ahead of time, recognizing that common mistake. Great games and examples for that resource! Thx.
Ps. I don't think Maia will dethrone the King, but just give us mortals a chance at mirroring him -- and making something great of our own play -- or mirror other greats too! He will continue to improve too I think, and is maybe the future goat, as they say -- the best ever! Thx again!
Left before rematch - The Most Human Move.
The next update will have Maia call you a bot and block you when it loses.
24:30 “I forgot the knight could even take like that” - Jerry the Beginner (2568)
33:38 He forgot the knight could even take like that, lol
Very nice games, fun to watch... I believe you missed something at 32:30 Nf6+ fork winning blacks queen and ending the game... as the bishop is pinned
Nf6 FORK AT 32:27 ??
There was a missed check 32:29 and maybe mate after.
The best as usual, Jerry! Keep going
So cool and exciting. Love that there are all open source projects.
32:37 ouch that's human-like...first player blunders then second player doesn't notice and blunders right back
8:23 missed rook and no excuse for a lack of time.
I love the content Jerry! This is so cool! I feel like next time I’d like to see you play without the urgent time control so you have time to think and talk out your moves though. Your analysis and critical thinking always amazes me.
32:30 You had a tactic there :)
Yep. Free Queen!
Even in the final position. QxB QxQ Ng5+ forks the K and Q.
I also want to see Maia chess try and flag opponents when in losing positions. Also it should play troll like openning moves when facing opponents whose elo is significantly lower. This is what Carlsen and high level players do on Lichess.
I recommend playing against lichess BOT "Boris Trapsky" - he plays well like a 2100 but creates unsound traps for an oponent :). The opening phase he plays is from a set of most played moves at lichess in that position :)
Amazing! Love the videos jerry. Never stop
32:30 The knight was forking the king and queen :(
Even better at @32:38
She doesn't stop to think like a human though... That's definitely a big edge with faster time controls
The chess cheats are gonna be weaponized with covert cheats with this.
Thats actually a really scary thought. Didn't even consider that. Only seemed like a really good learning tool to me. Yikes
actually if people really wanted to cheat they can still play suboptimal human like plays with stockfish suggestions. Kind of a waste of time though, I do think as a result more tournaments should happen in person between a chess board.
@@primekrunkergamer188 it's hard to believe chess cheats enjoy being nothing more than a surrogate for a chess engine but they are out there. Guess it helps them feel better about themselves for being absolutely trash at the game
People already are successfully cheating... This will be easier probably, but still
Well, this is probably not so human, in a bullet game when either sides time nears zero the bot is speeding up like hell and still finding good or only moves.
Amazing Job! Never Stop! We appreciate you so much!
I love Jerry's videos.
Playing against the Maia's is a sentence I usually only hear in Age Of Empires 2 commentary.
adding time to your clock each game is not very human
Maia learned from 12 million human games played online. And out of those 12 million games 10 million of them were cheating with an engine. So Maia is the same as Leelazero and Alphazero at the end of the day.
Lol
Starting an online petition to rename the bots to maia.hee, maia.ha, maia.ho, and maia.haha
jesus fuck i havent heard that in years, thank you.
Thank you @Pizzous, you made my day! :)
To Bee or not to be, maïa
At 32:26 I got excited for Nf6+ winning the queen >:)
Maybe the moves are more human, but the way it plays, spending no time at all, is so inhuman and infuriating. Nobody will ever blitz out moves like that in 3+2, that's one of the main reasons I don't like bots. They should include some kind of waiting time after the calculation is over.
You are right. I prefer to play without time against bots. I am pretty sure they could simulate the time easily too, but you would have to get a big DB with the times recorded, and I can't think of any.
@@sanferrera Does Lichess have a database that includes time spent? The time spent is right there with every game analysis
@@harleykf1 Wow. That would be great news. When I download games with the interface it doesn't show the times, but if you say they appear in the analysis, they probably have it, but don't send it when you ask for the pgn. That would be an amazing source of info for the Maia team, if they wanted to go in that direction.
Who needs sleep when I can watch chess.. Always willing to make the exception when Jerry uploads :D
26:24 free bishop if rf1 right?
I hope this will never go public. The last thing we need is an engine that is difficult to detect.
Good point
@@alfa-psi dude he's saying it should not go "public", of course it is a great scientific achievement and should be used to improve our understanding of the game. He's just conveying the concern about online cheating which is a serious problem.
For now it's only 1750 blitz strength as a style matching alg like this is incapable of searching more than one move deep (since it has no eval function)
how about maia lined up against all the games of past champs...fischer_tal_morphy _capablanca_aljekhin, etc
32:26 should've forked the Q after 24... Qh5
Maia moves instantaneously. No human plays like this. They need to add a human-like time element too.
Why, so you can wait around doing nothing?
@@Joe-tx6vc 16:03 because you're also playing against the clock. You should be thinking during your opponents turn.
It would be fun to watch Maia chess learn to rage quit. This is the ultimate "human move"...
Or refuse a rematch
Moving instantly is far from human-like kind of playing LOL
I feel like you were a bit too intimated by the fact that you were playing an engine, missing that its distinctive feature is that it blunders a lot (at least these versions, that were trained on blitz games). For example, at 25:21, you can win an exchange with Nxd5, at 26:28, you can win immediately with Rf1+ and Ra8+, and at 32:30, Nf6+ also ends the game immediately.
Love when Jerry plays with the increment.
I tried a couple of games, but she just plays three moves and then logs out.
I've been waiting for developers to do this, make an engine that plays like a human and being able to intelligently handicap
No matter how many games Maia studies, it will never master the art of mouse slipping.
@5:40 I never understand why this pawn isn't hanging can someone please explain to me, I see grandmasters never take it but I don't understand why. I know the dragon is behind the f6 knight but I still don't see any discoveries because the queen is not on d2?????
Do you have the correct time stamp? Which pawn do you consider to be hanging?
Do you mean the e5 pawn by black? If you mean that, it is because after dxe5 dxe5 Nxe5 Black can play Nxe4! And now the material is equal and both knights are attacked, blacks and whites
@@Namelessuser501 Thanks so it basically is just a big exchange.
@@benjamincarter6095 e5.
The last game was pretty crazy! However, I almost jumped out of the chair when you didn't take on f6 immediately winning at 32:38 :P I guess great players also miss easy moves sometimes
In the final position of the last game white has Qxh3 Qxh3 Ng5+! winning another piece.
Good eye, wouldn't have seen that!
This is gonna be so useful when you don't have a chess partner to play with
Maia was never designed to play chess and chess is not its destiny.
I was so intrigued to read their website. Looking forward to seeing all the different ways it can help.
As it states on their web site: "Maia is an ongoing research project using chess as a case study for how to design better human-AI interactions."
You’re statement is misleading. It was directly designed with chess in mind, but they’re hoping to use the insights from this project for non-chess related activities.
@@Mecinimi Once you learn how to read, take as long as you need to read their website. Then keep re-reading it until you understand. If you still fail, maybe go and play in the sand pit.
@@PushyPawn Yikes. Where's all that hostility coming from?
@@sashaboydcom The year 2020.
32:29
"I'd love to get the queens off"
25. Rg1 Qh7??
great blunder. sadly not spotted with less than 5s on the clock
(..and again after 29... Kxg8?? )
It's interesting to see that it's easier to beat Maia 1500 vs beating Stockfish at 1500 strength
I made a blunder against maia1 (it was rated 1600+ when I played) that would have been a sure loss against Stockfish at that level. The funny thing is I made a comeback and actually got to a winning position up 2 pawns but then blew it with letting it nail me with a perpetual check/draw :( Very human like game though! (compared to Stockfish) EDIT: I'm one of those mostly play casual against the computer people........but I'd estimate I'm about 1500~1600. Best win was 1760ish human and level 6 (2300) Stockfish, but that SF lv6 was sooo mentally draining to beat for me lol.
@ChessNetwork i recommend playing against lichess BOT "Boris Trapsky" - he plays well like a 2100 but creates unsound traps for an oponent :). The opening phase he plays from a set of most played moves at lichess in that position :)
Thanks for the content. Beginner here trying to learns and your vids help me to know what to look for!
Great to read that 👍🏼
Wow, Imagine having Carlsen play against Maia Fischer
Jerry's back vs the Cyborgs. It's on!!!
Lol 16:56 Jerry predicts every possible move as “the move” Maia5 would make
What do you mean? He's saying that the computer will resolve the tension instead of not moving the pawn at all
Nf6 forking King and Queen after availing itself three more times for you to play and you ignore it 32:27 - 32:29 , 32:38
"Am i a joke to you"
Jerry proves he can play more like a human than Maia
I’m only rated around 1600 and I have two wins against Maia9. Both times Maia made blunders regarding discovered attacks. I feel like a human would evaluate these positions a little better. Maia plays too fast imo
A chess site that fearlessly shows his skills. Well done! 👍
Wow, those moves felt so human-like!
Missed queen at 32:28
@ktbDash He had 2 such chances. First @32:26 & then a move later at @32:29
this is awesome thanks Jerry
7:36 I should not have laughed so hard at that blunder. We know what you meant to do Jerry. No worries! lol
Surely there was Nf6 for a couple of moves, winning the Queen, at 32:28 onwards? Amazing what you can miss in time pressure.
Yup.
I have to say that something like this could the end of honest online chess. Because if this algorithm could be used or mimicked and perfected, then a person could use the maia engine to cheat and then the online chess sites might not be able to detect it. Its a great technology but its unfortunate if it ends up just making it easier for people to cheat.
Hmmm that's interesting, but although it's human-like, you wouldn't make 2000 moves with only 1600 account.. You'll eventually blunder, so maybe harder than stock fish or Leela to detect but it's probably obvious. It's like letting magnus carlsen play with your personal account, although it's human-like, you definitely cheat with magnus beside you... When you, let's say, are only 1800.
@@diskyariajetmiko They have versions from 1100 to 1900 though. It's very plausible to use a lower level version of the engine and be fairly undetected
So human to move instantly on every move and end up with a minute more on your clock than you started. Shouldn’t they make it pretend to think?
It would be unnecessary to make human wait for virtual "thinking"
It no longer moves instantly. It's been adjusted.
@@diskyariajetmiko well, the point is to make it human-like, and people think on the opponent move usually, you don't have time to think on your opponent turn if they play instantly. It's been modified anyways.
@@ChessNetwork ah thanks, and thanks for the great video
What if a AI could imitate Bobby Fischer's style??? Man, would be very cool if it happens...
or Tal. I feel like Petrosjan already exists in Komodo