ฝัง
- เผยแพร่เมื่อ 12 ก.ย. 2024
- Join NM Sam Copeland (@sam_copeland) as he breaks down 5 chess positions that chess engines don't understand. Despite the clear superiority of engines, there ARE positions which chess engines don't (and possibly can't) understand that are quite comprehensible for human players. Typically these positions showcase the human ability to think creatively and formulate plans and understand long-term factors in the position.
Read the full article for more positions chess engines don't understand:
www.chess.com/...
Follow us here 😀:
✅ Sign up for FREE online play: www.Chess.com
💜 Check us out on Twitch: / chess
📸 Follow us on Instagram: / wwwchesscom
📱 Like us on Facebook: / chess
🎶 Follow us on TikTok: / chess
💙 Follow us on Twitter: / chesscom
#chess #chessengines #computerchess
Last position is just beautiful
69 likes 👏
The only position I couldn't see a plan. Although I've analysed a couple of these positions before.
I found the plan in the first one, since it's basically the only plan and it just involves understanding that Qa2 is the only way to break the position, and conceptualising how to defend. Once you understand that black cannot take a rook, the rest is easy.
I am thankful for your likes.
the suffocate comment is right above this one for me, how coincidental lmaoo
That first position is making me suffocate.
Same here like hell, made me feel sick
like i genuinely felt uncomfortable looking at it
@Vishesh Gautam It's the exact opposite of satisfactory.
I spent so long seeing if en passant was a good move for black, but I cant calculate it
Quite literally
Position where chess engines are wrong: Bongcloud
Bongcloud is a psychological advantage not a positional one, so it's impossible for a computer to understand
It has to be played against an opponent who doesn't expect it
Catching them off guard is the advantage
@@dustinjames1268 how is there an advantage in exposing ur king immediately. White or black can just castle and then play for an attack by ramming his pawns.
@@dominusdone5023
It throws your opponent off their game
Not only does it delay your development leaving them with nothing to respond to (thus more likely to make mistakes that you can punish) but it also allows you to analyze the playstyle and strategy they're attempting before you develop
It's purely psychological. The position is obviously not in the favor of the bongcloud player, but in the hands of someone skilled like Hikaru or Bobby Fischer, it's a powerful opening.
Its origins are much earlier than Hikaru, he was just the one to popularize it
@@dustinjames1268 is it why the engine evaulates it as pretty much completely lost?
@@dustinjames1268 the people that actually play the opening and survive are high level gms like hikaru. You would get crushed normally if your opponent knows what to do.
How can't Stockfish, one of the strongest chess engines in the world, better than any man or woman, able to see mates in 72 moves, analyse a closed position where all you need to do is move your king around, take the pawn not the bishop or just think about underpromotion
It probably doesn't see it at low depth.
I ran the first scenario through Stockfish yesterday and it didn't exchange the pawn for a rook. It did take about 5 minutes of analysis, though, to see the correct move that leads to mate.
Only Alpha Zero can.
Move your king around? That alone is 8 possible moves per depth. Number of possible moves grow like crazy
@@rk2280 No, i tried it with stockfish and it can start using the queen at 16th move. But i had to go a bit deeper in order to find the solution in 16 moves. Not much deeper tho, only 0.5 seconds and just in few moves.
This means to beat a chess engine we must get to one of these endgames
Winning a game by memory is not beating an engine. You could just memorize a perfect chain of moves and since the engine will always play the same moves you would always win but try beating it on fisher random. Gl with that.
Not even Magnus could last until endgame with the latest Stockfish, Commodo or really whatever modern up-to-date engine you choose.
@@VylinJanKranBlanMichal latest? Lol even deep blue from 1997 would probably beat Magnus.
@@siddhantjhaveri so a newer release would do it with an even greater ease, wouldn't it?
@@VylinJanKranBlanMichal obviously. What I meant was chess engines were comparable with the best humans more than 2 decades ago. Now any chess engine will absolutely obliterate any human.
The fact that Stockfish dev solves most of these puzzles, and the rest it solves weakly (evaluation is wrong, but the moves are correct) just shows how incredible state of the art engine it is.
First position is normal when you are a beginner
Underrated comment
Hindi naman
That last puzzle position is a thing of beauty
If you’re wondering about gxf2 at 14:14, white plays g3+ and both Kxg3 and Kh3 are stalemate because of the pawn that just captured the bishop
Thank you! I was wondering why that ended up as stalemate too, but it makes sense as it's the same idea as the final queen sac
Crystal, a chess engine forked from Stockfish understands these positions
How does it work?
@@yit6 Well, I don't exactly know everything about it. But what I understand is that it prunes less aggressively. There's probably more into it
@@swollpenispok8172 Yes it does. It's a chess engine specifically made to detect fortresses and solve difficult positions
@@swollpenispok8172 Obviously not how it works. If you still don't believe me, you can go search "Crystal chess engine wiki". You can also find the GitHub page there
@@swollpenispok8172 no. Not how things work.
Stockfish 13 found all the right moves with little to no effort.
Yes! I'm fun at parties, leave me alone! :c
you must be fun at parties
Sometimes these famous positions are added to engines and this is why they know the solution
its about the evalutions, he said in the video the engine will say black/white is better not that it wont find the best moves.
Checking the data is sexy change my mind
StockFish 14 in depth 40 says -5.6 advantage for black in 5th position. And as a 1400 rated played i just realized it is draw.
Well, I took the time to download the latest version of Stockfish and let it run through these positions and, as surprising as it is, it did come up with the right ideas in every puzzle except for the first one (I was too lazy to wait for it). In some puzzles it is true that it evaluated the position as losing, or with an advantage for black, when it was a draw, but even in those cases its best move was the one showed in the video
the engine also evolve
Stockfish had a stroke calculating this
me too.
Then stockfish choked and died. Rip Stockfish
not true, just download stockfish 14 and try it on your own.
@Utsav Joshi No it doesn't. The only way stockfissh would know this is because some famous positions are programmed into engines.
The first position is so beautiful! I admit that even though I've discover so many positions that engines cannot solve, this one is astonishing. Thanks Sam!
It makes me feel claustrophobic!
@@leemarshal3329 i would have felt claustrophobic if white couldn't have won it. But White did win.
I tested all of those positions and SF 14 was able to come up with all of them in a couple of seconds
Of course.
But most people will not factcheck these things.
Stockfish: Best chess engine
Also stockfish:
Crystal: My moment to shine... 😂
@@fahimp3 lol
@@Helmutiii Actually I just checked with my laptop's SF13 (with 5-man TB) and it found the correct moves to all except the Giri game where it took a more complicated approach (still won). For 2 of them even though it found all the right moves it's eval was wrong at the beginning for many moves...
@@fahimp3 that's wired😂
@@Helmutiii But I don't think this video is clickbait since correct eval matters when it comes to the engine deciding to go for the starting position in the first place...
That Giri game is amazing.
2:51 i feel so claustrophobic just watching it
I like these studies a lot.Hoping more videos come out about the topic
I know I’m making progress when seeing some of these positions makes me feel actually claustrophobic
Wow I never thought that a bishop promotion would come in handy at any position!!
Amazing solution for the 4th puzzle 😲😲😲
I once saw a mate in 87 or something crazy like that puzzle, which was completely forced, and not that hard to find, but ofc undetectable for an engine.
Engines can go to depth 100 if you give them enough time, offcourse our pc's wont, but an engine can still solve it.
@@bumboyy if the opponent move, mate is 1 move closer since its forced
@@bumboyy after 50 moves, it would be a draw, it pointless going any further
"Chess engines destroy the beauty of chess" - IM Levy Rozman
Then you have chess engines that calculate evaluation by piece development, king safety etc. But not by what move the opponent can play to counter that move so computer eval still kinda sucks
Crystal: Not so fast with your smug comment...
You’re right. Engines like AlphaZero and Lela really sap the beauty out of chess with their romantic, anti-materialistic play.
@@Corteum Viswanathan Anands famous quote goes something like, "for every door the engines have closed they've opened a new one."
Not true, in my opinion the engines play much more beautiful then human players.
We all know, that SF 14 would easely destroy tal, so why is tals play more beautiful when it is just worse?
the picture of a tournament briefly showed in this video brought back the feeling of the excitement I felt when I was attending chess tournaments during my younger years...
That first one was just asking too much. Stockfish normaly checks 18 moves into the future. You need to tell it to keep going and it should find it. Chess AIs aren't wrong about positions that require 20+ moves, they just arent allowed to look far enough ahead.
so what happens in the first position if black just does the french move? also black could just capture the rook once we have moved the pawn then he can capture the pawn as well after
(en passant by either pawn) Rxa2 bxa2 Kb2 and white is winning.
3:19 if you were wondering, Qb2+ doesnt work here because of the winning plan to get the other rook to b2 (Kd1, Re2, Ke1, Kf2, Re1, Rc1, Rc2, Rb2) to prepare b5 and infiltrate on the 4th rank. Giving up the b pawn first also works
I've been studying the hell out of the 1st position. With black putting up the fight by constantly blocking the pawn, white wades the rook between d1 and e1, it's a mate in 49. Eventually the black queen captures the rook, the white king captures back and the other rook has to be stepped by the pawn putting black in zugzwang again. With the horizon effect, it takes moves before Stockfish sees it's even a win for white, takes a while to see mate in 49. It sees mate in the 51st move, eventually on the 50th, then on the 49th. The white pawns move to allow the knights to break in the fortress and it's a win for white.
That last one is insane, really beautiful
There's engines playing like expert but not understanding closed position, there's GMs not playing perfectly but understanding these positions, and there's me, neither of these
Stockfish 13 understands position #1. Rooks on E1 and C2, King on C1, plays Qa2, which is captured, then b4. Depth 39, in 1 second:
1.Kd1 Kc6 2.Rg2 Kd5 3.Re2 Ke6 4.Rg1 Kf7 5.Ke1 Ke7 6.Kf2 Kf7 7.Re1 Ke6 8.Rg2 Kf6 9.Rc1 Ke7 10.Ke1 Kf7 11.Re2 Ke7 12.Kd1 Ke6 13.Re1 Kf6 14.Rc2 Ke7 15.Kc1 Kf6 16.Qa2 bxa2 17.b4 ...
Afterwards, Black sacs the Queen for the Rook, to capture the passed pawn. White announces Mate in 23 (which will likely improved if I let it continue to think over a few minutes).
2:14 you forgot pawn takes rock on g2 with a fork and then promotes to a queen.
That doesn't work because after Rhg1, regardless of whether or not the pawn promotes it will be captured by the rook and white will break through with the h-file open
My first impression on Position 1 is that you could play Rg2 and hope black takes but that is hope chess, the only real forcing line seems to be Qa2 and gobbling the pawns but it doesn't look like enough after black recaptures
elon was right chess is easy, kapp
8:15 promoting the a-pawn to queen or rook leads to stalemate as the black bishop is pinned and the black king has nowhere to go... i would rather underpromote to a bishop instead and force the black rook to defend the back-rank... but in this position, moves like b4 or Ra6 makes more sense than underpromoting the a-pawn to bishop or knight.
After b4 or Ra6, there's Rc8 and Ra8.
Can you imagine forgetting to play *b4* and *b5* before trapping black's queen in that first position? Black shuffling its king until it forces a 50-move-rule draw would be ATROCIOUS to sit through.
The last one is a masterpiece and is just unbelievable!
I always hated that stalemate was a draw but if it weren’t for that these sort of positions wouldn’t work
I am sorry, but in first position (fortress) there is still one chance for black what to play (so in time 3:47 it is not true that B pawn will deliver check mate so easily). You can stil play Qxb2 and at least prolong the game... b-pawn is not guarded any more and is also taken. And then there is little bit longer but still win for white using second rook...
correct
What happens in the first position after b4 by white if there is a en passant capture by either Pawn?
White forfeits
Holy hell
Thanks, I should have set it up and tried to work it out
White forfeits because en passant is too powerful
@@DaGaJbmKojJe ah, a member of anarchy chess
Hmmmmm stockfish 13 solve the first though I'm trying to do the rest later
4th posistion: i see no reson as to why taking the "mad piece" makes stalemate, you can promote the queen and then make a ram with rook and queen against the bishop on e8, only move that prevents instant mate is if king is on f8 before said ram is madebc the rook after taking bishop would stop king e7 while putting king into mate due to the pawns suffocating the king to the single rank king wouldnt be able to take rook either bc queen defends
After you take the "mad piece" it is an instant stalemate. You have no more moves to do anything you mentioned here
Let me add another very interesting position which is for computers impossible to understand because it appears as if it's not a blocked position, but even if in fact it isn't...in a way it is! --> 8/p3k3/2p5/1pPp4/3P4/PP1K2P1/3P1PBr/8 w - - 0 2 Here White only has a drawing move, that no engine finds, which is 1.Ke2!! with the idea that after 1...Rxg2 (of course whenever black plays ...b4, white replies with a4.) 2.Kf1 Rh2 3.Kg1 Rh8 (same is ...Rh7 Rh6 Rh5 or Rh3) 4.Kg2! followed by 5.f3!! and White has reached the key position where the Rook has no way to enter the White position, and it's a draw! Will engines one day be able to understand such finesses as this incredible one?
This position is wonderful. Who was the composer? I tested Stockfish from Lichess, it can't find the astonishing starting move Ke2 even after several minutes. After being forced to play 1.Ke2 Stockfish guesses the other moves of the line towards 5.f3 (even because they are very easy) but still it goes completely wrong with the evaluation, giving -3.8 and so a clear winning position for black.
thats honestly beautiful, an amazing position
Yea I found Ke2 instantly because I started to read your description qhile copying the fen in lichess and Ke2 is third choice of SF
What kind of profile pic is that 😂
Your photo profil is mad sus
I've tryed every position and my Stockfish 14 engine, on my Mobile Phone, did solve them all (after finding rhe right time control).
Sometimes very strong computer analysis engines are unable to win queen and minor piece against queen endgames with pawns on both sides in which the side that is a piece down has many checks. But to the very end of the game the engines persist in claiming the side that is a piece up is winning even when those engines are unable to win.
Giri's game is from candidates.
bishop d1. if game continues= possibly "double check mate".
black queen c7= double checkmate . ive found the most incredible draw ever. remove queen and f4. black pawn on black pawn on white. even if it looks like the black king is checkmated, the black king is actually not checkmated ...
In the 1st position why Enpassant with "a" pawn was not done
After b4 and either axb3/cxb3 is done (doesn't matter which) you get Rxa2, bxa3, Kb2 will stop the promotion of the queen, and white can now advance to either a4/c4 depending on which pawn captured enpassent, allowing them to move their knight to either a3 or c3 and allowing the rook to escape via the b file
In the first position, what if black sacrifices his queen for the rook and then king takes pawn
In your position from 1938, Stockfish 14 finds a8B at depth 61 (31st move) using ~6min on my hardware.
In the last position, your comment about engines thinking black is winning is not the correct way to think about it. For most engines, such as with Stockfish, it indicates the score is -4 (black winning) but that is only the measure of material. When an engine REALLY knows it is a draw, it will show a score of zero. But in order to get there, it needs to exhause every single one of the lines that lead to stalemate and there are millions. If you use an engine, you'll see that the moves they find are actually correct, they just don't have the time to prove it is a draw.
Good stuff though.
In the 2012 game, engines mostly like h4, but i didn't have time enough to check it deeply...will comment later!
I think the main problem for the engine is that they are not "connecting" the lines. Like when a human analyses the lines, they do the first moves and then understand at some point that they all lead to the same line they already analysed. Which is why it takes less time. The engine is treating each line as its own line I would assume and therefore, even if there is just a move difference, it calculates it as far as it can.
Number 5 is wrong, where bf2(!!) is played gxf2 allows kh2. So this is why the engine disagrees, because it’s right
So don't play Kh2 and lose, play g3+ and draw instead.
Title: Top 5 positions engines don't understand.
Me(a human):Also don't understands
Me:Maybe I am an Engine.
My brain: 🗿Stonks↗️
3:12 Do En Passant ! promoting queen is Simply a blunder
What happens if Black doesn't promote the pawn but instead takes the rook on c2?
3:12 it is important to not that if black takes the rook instead of queening white MUST move the king to b2 or all is lost.
In the first position, when the pawn is about to promote and black king cant take it, why not play Qxb2+? That way white must play Kxb2, and the black can capture the pawn before it promotes. After this, we have a stalemate again because white cannot move the remaining rook.
We need part 2!!🔥
last one was just wow!! i am a bit sad that i didnt pause and try to find that beauty by myself even if takes me days, all the other one i have solved but last one would be amazing if i saw that Bf2 after Qf8 defence, anyways thanks for amazing puzzles and i hope i will see some more puzzles like this in near future :)
The new Stockfish 15.1 immediately finds the draw on the last puzzle
My chess engine was capable of solving all these chess puzzles and said the same moves
6:35 is just amazing.
3:09 le pase is possible for black so he can protect pawn
Wonderful thanks Sam. It will be interesting to see _when_ Stockfish &co. can see these positions and be able to solve then.
Great content.
Thank you.
3:15 illegal move, a3xb3 or c3xb3 is forced by all rules of chess
Here me out it might be stupid.. but could black just sack the queen at 3:30 and then the black king just shuffles up and down, you could never get a rook to the b file because the knight can never move?
Not only that but the black king could take the pawn no problem
Ahhh it doesn't work, I just figured it out, rook would end up on c2 and if you took with the pawn then you'd sack all your pawns until u promote.. otherwise the rook gets to the file nice one
The perpetual check with the rook is evadable if you find a way to capture with your queen
Ran these on my system today using Stockfish 15 NNUE.
Puzzle 1 Fail (Other say it works for them perhaps i didnt run it long enough)
Puzzle 2 Fail
Puzzle 3 Fail (Depth 37 Still Wants H4)
Puzzle 4 Success Depth 41
Puzzle 5 Success Depth 32
Last one is great!
Make a video on Vishwanathan Anand's top 5 queen sacrifices.
3:15 what if black promotes to a knight though?
Me: *scratches head wondering how that first position even happened*
14:08 pawn g3 captures bishop on f2 is winning
It bothers me when things are just made for content. En passant on the first position is posible taking the B pawn and creating a chain again, isolating white’s A pawn. Right?
In the first example, what if black plays for a knight of a queen
3:10 why pawn f4 and not f3, why don't the opponent take en passant ? defending the pawn
For the first one room g1 if pwan takes h3 takes g4 takes f5 so on so forth with wight to promote to queen
3:13 mate ur waffling the move is an-passant a4xb3 it protects the pawn attacks the rook unsoubles the pawns on the a file and reinstates the amazing pawn structure if rook takes a2 then pawn takes and now u have to play kb2 to stop making a queen and the a pawn is going nowhare as you still cant move youre knight
White still wins there.
@@game_ender4317 no it dosnt
give me the line bro
If memory serves, Ken Regan posited in or about 2013 that a 3700 USCF rating represented perfect chess. If he was right or close then computer analysis engines are ridiculously overrated. He may indeed have been right or close. After all due to decreasing K factors, any current 3300+ strength computer analysis engine should easily score 100 percent against all humans while still falling far short of playing perfect chess.
On the first board, I'm a little unsure. If you shuffle the King a5 to b5, then ... bxa2, b4 axb. At this point white can't get a rook behind his passed pawn, and a4 Na3 can be met with ... a1=Q+, rb1 Qxa3+ ... I'm assuming there's a forced win there somewhere, but its not trivial...
What? King on b5 will not help.. cannot stay there.. black has to move and it will be the king who moves..
The only way to win the first position as white is if your opponent makes a misplay.
if Queen takes the pawn and pawn takes back the position is still a draw or incomplete.
actually, I think the reason the engine eval is wrong is only because the eval bar might assume that black is playing to win rather than draw and therefore views a loss and a draw equally as favorable and is comfortable playing into the win for white.
the implication is that chess engines are built for single use demos not for tourney play.
the engine is so competitive that it is cooperative, maybe.
Cooperatively is a whole biochemistry term btw and it kinda makes sense here
In the first game isn't black able to hold on by taking en passant with the pawn on a4 and then moving the a2 pawn ?
Sorry. The sequence suggested by me in my last comment does not work.
In the first position, why can't you enpassant with the A pawn. Wouldn't that force a closed position where white can never win?
After that en passant capture, white can free his position up because his own a-pawn isn't blocked anymore. For example, after axb3 en passant, white could play Rxa2. Then after ...bxa2 Kb2, no matter what black does next, white will eventually be able either push a4 and get his N out via a3, or get his remaining R out via the now open b-file, after which winning is trivial.
in the first position, why couldn't the black A or C pawns just en pessant the B pawn and be protecting the A2 pawn while attacking the rook? It'd be a much cleaner move for black to play, stop the past pawn and force black to move it's rook to the b file or surrender it, making black's victory nearly inevitable, no?
I think you bring the king on a1 and then sack the rook, bring the other rook and push the pawn (depends where he en passants
4:38 what if here instead of capturing the rook, play bb3 to block the white pawn from proceeding!
Like this White's trick doesn't work anymore and he will lose a piece and Black's rook still be free
wait in the first puzzle, what if they sack the Black Queen for the rook, now Black can just take the white pawn
what happens if en passant in first position?? my head hurts
Congratulations NM Sam Copeland I was very happy with this art or composition of chess, I will even post this theme and link on my page so that other people can access and see this magnificent masterpiece. Only tasks, commands of a chess machine, do not exceed human creativity. For a while
Problem 1 : Stockfish 10+WASM uderstand it by 1.Rg2
This first puzzle doesn't really work, black can sac his queen on b2 3:41 and then take the pawn and after that you can't sacrifice the rook because if you pass it on c2 dxc2 and after the king takes and black king does any move d4 won't get you anywhere...
But in that case you play d2-d3 instead of d2-d4 to make sure the position opens up ...
hold on stockfish 14 checkmated black in the first position
where can we get these positions if we want to try on the board
well, they’re right there
Sometimes you just guess the winning move easily. But then you have no plan.
What about the amazing chess puzzle that Tal solved? Why didn't you show that??
In this position, can't black just move the king back and forth, never taking any pieces in order to force a draw?
False stockfish find the answers to theses position
Great idea for a video! thanks
I recognize the last one but all the others are new to me. Very interesting positions. I know a few other puzzles that engines can't solve either. I even have a mate in 7 puzzle that stockfish can't see.
In the first person why didnt enpassant occur in response to b4?
5:44 why the white king didn't take the pawn???