UNBELIEVABLE!!!! Thanks so much for the post. I will cover the game for sure. Man there is so much going on so you can miss beautiful things like this.
Great analysis. The alphazero-based engines have learned things about position that have yet to be discovered by the SF-style engines. As a longtime contributor of CPU cycles to Leela it’s great to see her successes. :)
Stockfish 17 must have an algorithm somewhere that says "Do something risky rather than just shuffle around into a three-fold repetition when playing white"- otherwise hard to explain 59.f4
You're basically exactly right. Stockfish just trusts that its a better player than most opponents and refuses to concede a draw with the white pieces. Especially not when there was something in its code telling it it had an advantage when the position was actually just a draw (probably "I have a space advantage"). This flaw of mistaking a mere space advantage for something they "should" be able to convert even if they cant immediately find the best move is a flaw thats been in engines for a while, its exactly what Hikaru exploited to beat Rybka (one of the top engines before modern times where engines are now fully out of human reach).
To be clear, the "gaining space" strategy usually serves them well, we dont have infinite computer power so these engines never literally calculate all the way to the end. So even they have to follow, for lack of a better word, "intuition" once in a while. They wouldnt be so dominant if it didnt work 9 times out of 10, but it explains why they still make blunders like this.
Well, if your position is rated >1.0 and your choice is between repetetion (=0.0) and sacrificing a pawn (what yields to position_rating - 1 > 0.0), for a computer, it's obviously that even giving up a pawn for nothing is better than the repetition. Here, in all depth of the calculation tree, the sacrifice even gives white playing opportunities, so from white's point of view, it's not just a pure sacrifice. And seeing something that this sacrifice allows the opponent 25 or 30 moves later to bring the knight from a7 to g5 and creating three connecting passed pawns with a counter sacrifices, is just out of reach. But I guess, the problem here is the too optimistic position evaluation that leaded to the fatalistic approach trying to avoid repetition for a high cost, where white should have avoided this full blocked situation from the beginning (or just evaluate it to much closer to 0 than >1), what probably Leela does "intuitively" and then chooses with pieces switched with white a complete different strategy from the beginning. I wonder whether there's any rating of horizon problems in chess engines nowadays. Something like, if evaluation(depth=N) = s and evaluation(depth>N) < s what is quite typical if position is not as good as evaluated that as deeper you go calculating as worse evaluation you get, is adjusted somehow to the calculation, to avoid such kamikaze tactics.
The bigger issue is it evaluated itself as winning in a likely unwinnable position. If you evaluate the static position as +1 you will never choose the draw because 0 is a lower score. Because of that bad evaluation accepting the draw would have been seen as one of the worst moves to pick from. Engines optimize their evaluation functions to choose moves, if the evaluation is in error the decision that follows is suspect (literally pick the move that highest score however many moves in the future they can read). Normally you'd expect a move that loses something for not much to score worse than taking the draw, but there is something about the position that stockfish is over or underestimating and horizon effecting on.
Its evaluation probably went back from +1 to 0 in the normal lines, as the 50-move rule came near, so it decided to play a move that drastically altered the position and still gave it an advantage. It did activate the bishop, but it was very risky.
Hats off Jozarov todays ur analysis was infinite we can say chess has possesed with infinity no one can learnt chess within 1 birth you have to recarnate to learn it again & again 😅 .i ve no words I want to see the top most engines competing each other like komodo Dragon 3.3 vs Lc0 & SF17 and ALpha zero and houdini etc
This doesn't prove Stockfish is unbeatable. Logically, it only demonstrates that, if started off in a poor position, Stockfish can't work miracles. So, was the starting position poor? Apparently so, because Stockfish couldn't save it. How did the two compare playing from the same starting position, but with the sides reversed (ie, Stockfish playing black)? That result might show that Stockfish can be beat. But this contest was too lopsided to really show us anything definite.
Blocked positions are basically game of chicken. Whoever tries to unblock has to sacrifice something. Stockfish lost patience with 59. f4, which ultimately has led to his defeat.
@@Josh-cz3ymnope. Anthropomorphizing is not strictly wrong, as Stockfish is hand-coded by humans, and they program it to have "contempt" for its opponents, and sometimes play inferior moves to avoid draws by repetition.
@@jeddylajos when engines start "high level shuffling" pieces back and forth on their side of the board without making any improvement or progress, it means they don't see any moves that improve their position, but they must move ("zugzwang") so they make the "least bad" moves which change their positions by the smallest amount by their evaluation criteria. After shuffling for a while, they start approaching the danger of getting a draw (by 3 time position repetition or 50 move rule) so then their "contempt" programming kicks in: if contempt is set low, then they go ahead and keep shuffling to get a draw; if their contempt is set high, then they make the least bad move which avoids the draw, which is usually a pawn sacrifice (which always resets both the repetition and 50 move rule counters) and that usually gives an advantage to the opponent, often enough to win. So high contempt makes them intentionally lose an otherwise drawn position. I think Stockfish did exactly that with the f4 pawn move, as TCEC engines get their contempt set high to avoid draws.
Giving the engines pre opening moves will always leave either white or black with small slight advantage dependent on the line.. but programming the engines to always push a pawn to avoid 50 move draw is detrimental.. that F4 move made no sense and had no preparation.
So the main difference is that Leela knew the piece sac was very dangerous, while Stockfish just enjoyed his position and aimlessly shuffled his pieces. It would be interesting if Leela would be able to win without the piece sac.
I think this is the first time that it lost on both sides. Alpha Zero probably Had also Had some wins Like this against the older versions of Stockfish when it was without nnue
Wow Stockfish making desperate pawn moves really seems like the game was decided early the fish should’ve played for the draw its offense was stymied great game by Leela if Stockfish picked its own opening I’d bet on the fish winning the moves in this game so engine like no way humans playing like this funny trick Stockfish pull at end offering rook to get Queen Leela knows Thanks Josip
since stockfish had the positional advantage, it obviously calculated that draw by repetition was worse then sacrificing pawn while maintaining the positional advantage. why do you guys act like its behaviour was some mistery?
And The f4 move pawn sacrifice move, my mobile sf dont even show this f4 move in first 3 best preffered move, and this sf17 jz played it with risky sac! And after playing this move evaluation drop from +1.7 to +1.1 for sf himself! It shows that theres error in move choicing of sf17!
Chess is problem solving. Previous stockfish versions prioritize the most simple and quickest solutions. Stockfish 17 solves it different. So It has problems with time management. Update to 17.1 might solve problem or might not solve problem.
A fantastic win for Leela, you have indeed spotted a blocked out position problem for the fish. This combined with such a high evaluation of the blockaded position for white made the acceptance of a draw too difficult. Thanks Josip.
I wonder how some of the other strong machines played this for white and black? Before The fish did bg5 it calculated it had a winning board, and then lose wow. Stockfish creators will have to really fix this hole. It could even be the perfect testing ground for future stockfish versions. iam analysing with sf17 the move where leela did bg5 and at 37 ply sf17 now moves Kg1. At 50 ply Kg2 so thats prob not tot losing. So the mistake seems to be somewhere in the range of the 40's ply depth.
SF 16 was arguably worse at positions like this... from observation, SF avoids these positions as white and excels at defeating engines as black whenever it gets to play fortress style positions because it evaluates the position as not an ideal setup for either side. However, The difference in search functions for these two engines is highlighted with experience and favors Leela.It's tough to linearly think your way out of some of these positions. The neural net is what I think the wekpoint for this version is atm. They haven't updated the default net using SF17 engine yet. Something they normally do within the same week of release... the default net in use right now is from 1 month ago.
@@chrisdavis2161 I think you are right, at higher time setings SF17 does not play bishop C2 but plays Kg1 or Kg2. You think they still will update the default net?
@@blijebij they make a new neural net almost everyday, however, the default net has to pass a certain criteria for it to be made into default (I'm assuming this by the way... I'm not a programmer... I'm just a computer chess engine enthusiast that became increasingly interested in AI chess programming about 5-6 years ago). However, to go back to your original comment... I do not think that the depth matters for which line SF will choose as much as the amount of nodes it analyzes and which branches get pruned for bad continuations. However one thing to keep in mind is that these positions that they make the engines play are bad (Some are really bad). I believe that Leela will always have the advantage when it comes to playing bad positions because logically, it has played more bad positions than any other engine, because that's how it learned how to play chess and how to differentiate good and bad moves. SF pruning bad continuations from a bad opening has more of a burden of proving its moves are correct over time.
The usual chess practice looks different, you have little advantage from opening, even with ENG server games. Therefore, Stockfish dev can create 100 UHO Elo more, in the current practice is not to see any of it! The disadvantage of too much selectivity is seen in analyses! There are positions where SF 17 is simply blind due to fast search speed (which is good for UHO) and takes a long time to see the winning move. Less UHO and less selectivity would be good for Stockfish, but the practice is different, the UHO-Elo madness knows no limits. Daily Stockfish dev produces new Elos (you can observe on Fishtest daily, there are counted as Elos peas and are sacred so to speak). Well then, continue so, cheers to the next UHO-Elos! :-)
Lost count of how many times you said 'Blocked Out'. I have never heard such an expression used in English before, to describe a chess position. Native English speakers say simply that a position is 'Blocked' or 'Blocked Up'.
Ok I took you to far for 17 for you to understand for it is and all included for all reasons I say of the situation in multi situations with all of it's processes for things to work in these rules of engagement so here's another from the chief and head programmer,when in how in the beginning where the results are the answers no matter what actions are taken in the procedures any failure is from the lack of knowing the process of examination.😮
What makes Stockfish 17 46 Elo stronger than Stockfish 16 ? Answer: In the own tests performed by the developers, mostly in the mode 10s + 0.6s and 60s + 0.6s bullet games. In addition, the vast majority of games were played with UHO (unbalanced human openings). One side has a great advantage here, but it is difficult to realize on the board between people. The advantage of Stockfish 17 or the newer developer version is that SF dev is optimized for such variants. And what’s wrong with that ?...
Well, it's that first of all in fast games the time is higher than in the tests that are done with SF by its programmers, and second of all, that, he is ”specialized” on UHO tests and by those no longer finds the resolution to the positions that appear. SF programmers have made many changes to his detriment since may... You can no longer trust the analysis that you do with more time. In my opinion, there will be more defeats. So maybe the programmers will wake up - to see that in the race for ELO on very short tests and prearranged positions with great advantage - they managed to ruin SF for longer analysis.
I guess this is the part about cheating but there is many processs and procedures to accomplish the same thing for the same result so let the chief and head programmer show you a true insight,no matter how you replicate a situation you have to learn how to beat yourself to make the variation to have the continuation.😮
I actually checked whole game with my mobile stockfish, so i found out that this sf17 actually not choosing the best move that my mobile sf shows. This sf 17 actully prefers 2nd or 3rd prefered move or not from first 3 preferred moves even!!! So i think this sf17 must have some miscalculative glitch in there, and they will fix it in update. I think so!
@@Saata-ev2md then why power cpu plays f4 pawn sacrifice…which is blunder! Even jozarov says its miscalculation and normal sf says its blunder! Its glitch because this sf missing winning ideas, drawing possibilities many times!!!
Everyone knows SF17 is crap. The only question is did the Devs intentionally make it crap so the TH-camrs could spend the next 4 months spamming SF17 lost!!! Vids, or were they so dumb they forgot how to get AI to improve itself?
You understand if the engines play what they want you would see millions of berlin defenses and morphy defenses. What's the point of that then. Always the same comments
Doesn't Leela only use Nnue too? And this problem, if I'm not mistaken, stockfish had way before they removed hand craft evaluation. (Since it's lost this way to Leela a few times before)
I dislike Chessbots being displayed as some kind of super smart AI. Stockfish or Leela can't speak, can't think. The only thing they can do is play chess.
He spent the entire video complementing leela just about as much as possible. Stockfish being better is just the objective overall result, not his opinion
I love that you show were the games deviated and what Leela played in that position, it's really instructive!
Enjoy these computer chess videos. Keep it up!
This was definitely an engine game unbelievable how both were searching for good attack moves a lot of jockeying for position
You should take a look at game 81, Stockfish missed a stalemate tactic in a +4 position
UNBELIEVABLE!!!! Thanks so much for the post. I will cover the game for sure. Man there is so much going on so you can miss beautiful things like this.
@@Jozarovschesschannel Game 204 is also interesting. Leela beat Stockfish in 49 moves, because Stockfish missed a mate in 10.
New stockfish is weaker ?
@@nameno1230 hahahaha somehow that sounds really funny "missed mate in 10" XD
@@JozarovschesschannelCan you please show us this game?
Great analysis. The alphazero-based engines have learned things about position that have yet to be discovered by the SF-style engines.
As a longtime contributor of CPU cycles to Leela it’s great to see her successes. :)
Stockfish 17 must have an algorithm somewhere that says "Do something risky rather than just shuffle around into a three-fold repetition when playing white"- otherwise hard to explain 59.f4
You're basically exactly right. Stockfish just trusts that its a better player than most opponents and refuses to concede a draw with the white pieces. Especially not when there was something in its code telling it it had an advantage when the position was actually just a draw (probably "I have a space advantage"). This flaw of mistaking a mere space advantage for something they "should" be able to convert even if they cant immediately find the best move is a flaw thats been in engines for a while, its exactly what Hikaru exploited to beat Rybka (one of the top engines before modern times where engines are now fully out of human reach).
To be clear, the "gaining space" strategy usually serves them well, we dont have infinite computer power so these engines never literally calculate all the way to the end. So even they have to follow, for lack of a better word, "intuition" once in a while. They wouldnt be so dominant if it didnt work 9 times out of 10, but it explains why they still make blunders like this.
Well, if your position is rated >1.0 and your choice is between repetetion (=0.0) and sacrificing a pawn (what yields to position_rating - 1 > 0.0), for a computer, it's obviously that even giving up a pawn for nothing is better than the repetition. Here, in all depth of the calculation tree, the sacrifice even gives white playing opportunities, so from white's point of view, it's not just a pure sacrifice. And seeing something that this sacrifice allows the opponent 25 or 30 moves later to bring the knight from a7 to g5 and creating three connecting passed pawns with a counter sacrifices, is just out of reach. But I guess, the problem here is the too optimistic position evaluation that leaded to the fatalistic approach trying to avoid repetition for a high cost, where white should have avoided this full blocked situation from the beginning (or just evaluate it to much closer to 0 than >1), what probably Leela does "intuitively" and then chooses with pieces switched with white a complete different strategy from the beginning.
I wonder whether there's any rating of horizon problems in chess engines nowadays. Something like, if evaluation(depth=N) = s and evaluation(depth>N) < s what is quite typical if position is not as good as evaluated that as deeper you go calculating as worse evaluation you get, is adjusted somehow to the calculation, to avoid such kamikaze tactics.
The bigger issue is it evaluated itself as winning in a likely unwinnable position. If you evaluate the static position as +1 you will never choose the draw because 0 is a lower score. Because of that bad evaluation accepting the draw would have been seen as one of the worst moves to pick from. Engines optimize their evaluation functions to choose moves, if the evaluation is in error the decision that follows is suspect (literally pick the move that highest score however many moves in the future they can read).
Normally you'd expect a move that loses something for not much to score worse than taking the draw, but there is something about the position that stockfish is over or underestimating and horizon effecting on.
@wallysullivan9315 unfortunately you are way out of your depth here lil bro
I wonder if f4 was the losing move. If it was, then Stockfish 17 couldn't see far enough in the future to realize it was losing. Never play f4!
Its evaluation probably went back from +1 to 0 in the normal lines, as the 50-move rule came near, so it decided to play a move that drastically altered the position and still gave it an advantage. It did activate the bishop, but it was very risky.
Ben Finegold was right all along.
Love these videos man. Keep em comin!
I love the way you explain things. Youre so kind.
Thanks for the good words.
Thank you 🙏
A very brilliant choice by you, what a game!
👍
Maybe Stockfish 17 was running out of time but this is no excuse because both engines became the same time control.
Or maybe the selected opening was bad for white on the engine strength level
@@SecureHandle no the opening for white was good
@@anonym5160 eh not really. Plus some “good” lines or defenses are a setup for losing on engine level. I recommend you check it in the engine maybe
@SecureHandle Leela won in the same position with white against stockfish.
The fact is, stockfish just wasn't good in this set up
Hats off Jozarov todays ur analysis was infinite we can say chess has possesed with infinity no one can learnt chess within 1 birth you have to recarnate to learn it again & again 😅 .i ve no words I want to see the top most engines competing each other like komodo Dragon 3.3 vs Lc0 & SF17 and ALpha zero and houdini etc
Yes Jozarov - these are closed positions and they would be the only ones
where perhaps a human could beat Stockfish.
Great game.
10:53 => this is called "high level shuffling"
Incredible game
This doesn't prove Stockfish is unbeatable. Logically, it only demonstrates that, if started off in a poor position, Stockfish can't work miracles. So, was the starting position poor? Apparently so, because Stockfish couldn't save it. How did the two compare playing from the same starting position, but with the sides reversed (ie, Stockfish playing black)? That result might show that Stockfish can be beat. But this contest was too lopsided to really show us anything definite.
Stockfish is mortal!
Where are those guys who were saying "Leela is a joke" lol
Blocked positions are basically game of chicken. Whoever tries to unblock has to sacrifice something. Stockfish lost patience with 59. f4, which ultimately has led to his defeat.
No need to anthropomorphize. Stockfish evaluated that move the highest, so it played it.
@@Josh-cz3ymnope. Anthropomorphizing is not strictly wrong, as Stockfish is hand-coded by humans, and they program it to have "contempt" for its opponents, and sometimes play inferior moves to avoid draws by repetition.
I'd bet it's not the case. Maybe elaborate on this one?
@@jeddylajos when engines start "high level shuffling" pieces back and forth on their side of the board without making any improvement or progress, it means they don't see any moves that improve their position, but they must move ("zugzwang") so they make the "least bad" moves which change their positions by the smallest amount by their evaluation criteria. After shuffling for a while, they start approaching the danger of getting a draw (by 3 time position repetition or 50 move rule) so then their "contempt" programming kicks in: if contempt is set low, then they go ahead and keep shuffling to get a draw; if their contempt is set high, then they make the least bad move which avoids the draw, which is usually a pawn sacrifice (which always resets both the repetition and 50 move rule counters) and that usually gives an advantage to the opponent, often enough to win. So high contempt makes them intentionally lose an otherwise drawn position.
I think Stockfish did exactly that with the f4 pawn move, as TCEC engines get their contempt set high to avoid draws.
@@illarionbykov7401 Thank you.
Alapin is not good from +0,2 to 0,00 but it‘s surprising that Stockfish 17 lost!
Nevermind the preranged opening has nothing to do with the alapin.
Idk why you showed the evaluation for the alapin - all openings when played perfectly are 0.00. But yeah, this wasn't an alapin game
I am wondering if this is the same leela that lost to stockfish 16.1 or it's a new version?
Why did Stockfish never vlog about this?
This is the worst Timing for SF LOL, right after release!
Lc0 🔝🥊😵💫😁
My SF17 doesn't show 59. f4 at all? How it happened?
Giving the engines pre opening moves will always leave either white or black with small slight advantage dependent on the line.. but programming the engines to always push a pawn to avoid 50 move draw is detrimental.. that F4 move made no sense and had no preparation.
So the main difference is that Leela knew the piece sac was very dangerous, while Stockfish just enjoyed his position and aimlessly shuffled his pieces. It would be interesting if Leela would be able to win without the piece sac.
Hans is ditching his stockfish butt plug now he can't trust it anymore
Come on Fish. :(
Why you always get defeated by Leela🤦♂
I want yo see stockfish 16 (white) vs stockfish 17 (black)
Me too, great idea!
Most likely draw as always, ELO difference is not high
Has it happened before for stockfish to lose same opening as both black n white?
I think this is the first time that it lost on both sides. Alpha Zero probably Had also Had some wins Like this against the older versions of Stockfish when it was without nnue
There is at season 15 superfinals at TCEC at game 61-62 in Trompovsky attack
Game 204 is also interesting. Leela beat Stockfish in 49 moves, because Stockfish missed a mate in 10.
Today I Had 4 great suggestions by viewers of some top ai games. We have work to do 😁
@@Jozarovschesschannel For sure! This final is much more interesting and dramatic than the previous ones
Wow Stockfish making desperate pawn moves really seems like the game was decided early the fish should’ve played for the draw its offense was stymied great game by Leela if Stockfish picked its own opening I’d bet on the fish winning the moves in this game so engine like no way humans playing like this funny trick Stockfish pull at end offering rook to get Queen Leela knows Thanks Josip
Why did Stockfish make such a senseless move?
since stockfish had the positional advantage, it obviously calculated that draw by repetition was worse then sacrificing pawn while maintaining the positional advantage. why do you guys act like its behaviour was some mistery?
And The f4 move pawn sacrifice move, my mobile sf dont even show this f4 move in first 3 best preffered move, and this sf17 jz played it with risky sac! And after playing this move evaluation drop from +1.7 to +1.1 for sf himself! It shows that theres error in move choicing of sf17!
OMG 128 moves
Chess is problem solving. Previous stockfish versions prioritize the most simple and quickest solutions. Stockfish 17 solves it different. So It has problems with time management. Update to 17.1 might solve problem or might not solve problem.
A fantastic win for Leela, you have indeed spotted a blocked out position problem for the fish. This combined with such a high evaluation of the blockaded position for white made the acceptance of a draw too difficult. Thanks Josip.
Stockfish actually bleeds ...
Nice to know SF is still fallible
I wonder how some of the other strong machines played this for white and black? Before The fish did bg5 it calculated it had a winning board, and then lose wow. Stockfish creators will have to really fix this hole. It could even be the perfect testing ground for future stockfish versions.
iam analysing with sf17 the move where leela did bg5 and at 37 ply sf17 now moves Kg1. At 50 ply Kg2 so thats prob not tot losing. So the mistake seems to be somewhere in the range of the 40's ply depth.
SF 16 was arguably worse at positions like this... from observation, SF avoids these positions as white and excels at defeating engines as black whenever it gets to play fortress style positions because it evaluates the position as not an ideal setup for either side. However, The difference in search functions for these two engines is highlighted with experience and favors Leela.It's tough to linearly think your way out of some of these positions. The neural net is what I think the wekpoint for this version is atm.
They haven't updated the default net using SF17 engine yet. Something they normally do within the same week of release... the default net in use right now is from 1 month ago.
@@chrisdavis2161 I think you are right, at higher time setings SF17 does not play bishop C2 but plays Kg1 or Kg2.
You think they still will update the default net?
@@blijebij they make a new neural net almost everyday, however, the default net has to pass a certain criteria for it to be made into default (I'm assuming this by the way... I'm not a programmer... I'm just a computer chess engine enthusiast that became increasingly interested in AI chess programming about 5-6 years ago). However, to go back to your original comment... I do not think that the depth matters for which line SF will choose as much as the amount of nodes it analyzes and which branches get pruned for bad continuations. However one thing to keep in mind is that these positions that they make the engines play are bad (Some are really bad). I believe that Leela will always have the advantage when it comes to playing bad positions because logically, it has played more bad positions than any other engine, because that's how it learned how to play chess and how to differentiate good and bad moves. SF pruning bad continuations from a bad opening has more of a burden of proving its moves are correct over time.
Can you please make a Stockfish Gigachat Moments Complication?
The usual chess practice looks different, you have little advantage from opening, even with ENG server games. Therefore, Stockfish dev can create 100 UHO Elo more, in the current practice is not to see any of it!
The disadvantage of too much selectivity is seen in analyses! There are positions where SF 17 is simply blind due to fast search speed (which is good for UHO) and takes a long time to see the winning move. Less UHO and less selectivity would be good for Stockfish, but the practice is different, the UHO-Elo madness knows no limits. Daily Stockfish dev produces new Elos (you can observe on Fishtest daily, there are counted as Elos peas and are sacred so to speak). Well then, continue so, cheers to the next UHO-Elos! :-)
Jozarov, it would be interesting if you analyse this game with Stockfish 11 to see if the problem is in the old code or if it is in the NNUE code. 😁
Cool idea
WHAT ????😳
Lost count of how many times you said 'Blocked Out'. I have never heard such an expression used in English before, to describe a chess position. Native English speakers say simply that a position is 'Blocked' or 'Blocked Up'.
Ok I took you to far for 17 for you to understand for it is and all included for all reasons I say of the situation in multi situations with all of it's processes for things to work in these rules of engagement so here's another from the chief and head programmer,when in how in the beginning where the results are the answers no matter what actions are taken in the procedures any failure is from the lack of knowing the process of examination.😮
Failing to generate plans in blocked positions, trapping its own pieces... It feels like 2018 all over again.
What makes Stockfish 17 46 Elo stronger than Stockfish 16 ?
Answer: In the own tests performed by the developers, mostly in the mode 10s + 0.6s and 60s + 0.6s bullet games.
In addition, the vast majority of games were played with UHO (unbalanced human openings). One side has a great advantage here, but it is difficult to realize on the board between people.
The advantage of Stockfish 17 or the newer developer version is that SF dev is optimized for such variants. And what’s wrong with that ?...
Well, it's that first of all in fast games the time is higher than in the tests that are done with SF by its programmers, and second of all, that, he is ”specialized” on UHO tests and by those no longer finds the resolution to the positions that appear.
SF programmers have made many changes to his detriment since may...
You can no longer trust the analysis that you do with more time.
In my opinion, there will be more defeats.
So maybe the programmers will wake up - to see that in the race for ELO on very short tests and prearranged positions with great advantage - they managed to ruin SF for longer analysis.
They realease a new version Sf 3 days ago saying "Fix net downloading script" they said its a bug so maybe thats why
That's an unrelated nonperformance issue
That was a fix for a change made after SF 17 was released. See "Export and clean up net downloading script".
Stockfish 17 need to be fixed
I guess this is the part about cheating but there is many processs and procedures to accomplish the same thing for the same result so let the chief and head programmer show you a true insight,no matter how you replicate a situation you have to learn how to beat yourself to make the variation to have the continuation.😮
How to download stockfish 17 in cp.?
Что то хрень белые наиграли с пешками с3 и ф3, у меня так 1й класс играет))
I actually checked whole game with my mobile stockfish, so i found out that this sf17 actually not choosing the best move that my mobile sf shows. This sf 17 actully prefers 2nd or 3rd prefered move or not from first 3 preferred moves even!!! So i think this sf17 must have some miscalculative glitch in there, and they will fix it in update. I think so!
now question, do you think its glitch, or maybe your mobile sf17 is weaker than sf17 that runs on power cpu with multiple threads?
@@Saata-ev2md then why power cpu plays f4 pawn sacrifice…which is blunder! Even jozarov says its miscalculation and normal sf says its blunder! Its glitch because this sf missing winning ideas, drawing possibilities many times!!!
@@Ashtavakra_Gita cuz it had full big hash table and more cpu time to research and most likely decided that its the best move for him
@@Ashtavakra_Gita doesnt have to be a glitch, can just be wrong, its not playing perfect chess after all
12:43 SF evaluation bar showed "I've got enough"
Everyone knows SF17 is crap. The only question is did the Devs intentionally make it crap so the TH-camrs could spend the next 4 months spamming SF17 lost!!! Vids, or were they so dumb they forgot how to get AI to improve itself?
LCZero has been gradually catching up with stockfish in recent times.
What's so shocking, if the organizers intervene with pre-arranged openings.and moves ...
You understand if the engines play what they want you would see millions of berlin defenses and morphy defenses. What's the point of that then. Always the same comments
Leela also won the reverse position, so?
Sf 17 removed hcec hand craft evaluation and now totally relies on nnue that's why it's bad in closed positions
Doesn't Leela only use Nnue too? And this problem, if I'm not mistaken, stockfish had way before they removed hand craft evaluation. (Since it's lost this way to Leela a few times before)
@@Azariy0 leela doesn't have nnue it uses transformers . Sf 13 would have not lost this position because nnue was introduced starting from sf 14
@@ManojVijasf 13 had nnue, sf 12 was the one that introduced nnue and sf 11 was that last released version with classical evaluation
Leela is more poweful than stockfish 17!
I dislike Chessbots being displayed as some kind of super smart AI.
Stockfish or Leela can't speak, can't think. The only thing they can do is play chess.
Like only 1000 ELO better then the Goat Magnus!
64 squeres, 32 pices it is harder then anyone might think!
Ofc it's Lc0 who else
He loves hating on leela when everyone knows that leela is the best chess engine
leela is so good that it's lost the last 3 computer chess champions to stockfish
He spent the entire video complementing leela just about as much as possible. Stockfish being better is just the objective overall result, not his opinion