Further weaknesses

[Lyudmil Tsvetkov]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

[Uri Blass]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

engines are better than humans so I trust stockfish here.

I also think that there was a mistake of white inside the 35 plies when SF evaluated white as better so it is not a case when SF need to search additional 35 plies to see the truth.

I also do not believe that the gap between stockfish (after hours of search) and the ultimate truth is at least 500 elo

There are positions when engines are blind even after hours of search but they do not happen often in games.

[BeyondCritics]

Hi Lyudmil ,

Given the material signature alone (Queen against rook, bishop with connected passers on both side) it should be winning for the queen side in most cases, i believe.

But here it is not so an easy win as stockfish seems to believe. A passer with a protected rook behind it is very dangerous for white here, draining up all the queen activity. And white wins only due to its queen activity.


Oliver

[Lyudmil Tsvetkov]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

engines are better than humans so I trust stockfish here.

I also think that there was a mistake of white inside the 35 plies when SF evaluated white as better so it is not a case when SF need to search additional 35 plies to see the truth.

I also do not believe that the gap between stockfish (after hours of search) and the ultimate truth is at least 500 elo

There are positions when engines are blind even after hours of search but they do not happen often in games.

For the first time, I do not agree with a single thing you say.

1. Humans are in general better than engines, engines might have some tactical advantage, but humans are much better positionally.

2. It actually might need more than 70 plies here, as there is, at worst, a very very long series of queen checks, which prevents SF and other engines to make any reasonable sense of what goes on. For the time checks are given (maybe some 40-50 plies in a couple of series of checks) SF counts only the material advantage, does not see black promotion, so it sticks to its score. However, I have 2 reasons to believe black wins here:

- central passers are much faster in advancing than edge passers
- the pieces of lower power easily control squares both in front of the own and in front of the enemy passers, which is not true for the queen

3. I think the gap might be even bigger.

4. Positions where engines are blind even after hours of search do not happen often in games, simply because the engines know only a single style of play, an open one. And opening books are also devised accordingly. Really, I almost do not see closed positions in engine games, less than 0.1% for sure. However, when I play against SF, 95% of my games are closed. If I want to close the game, I simply close it and SF can not prevent this. So simply positions engines do not understand do not happen often in tests, because of the engine inability to play differently. But it is not only closed positions, there are at least 15 other totally different playing styles that have never happened in engine games and that engines will not understand even after a very long search.

[Lyudmil Tsvetkov]

Hi Lyudmil ,

Given the material signature alone (Queen against rook, bishop with connected passers on both side) it should be winning for the queen side in most cases, i believe.

But here it is not so an easy win as stockfish seems to believe. A passer with a protected rook behind it is very dangerous for white here, draining up all the queen activity. And white wins only due to its queen activity.


Oliver

Sorry really, too difficult for me to start analysing, it demands too many resources (you should not skip a single move), but I say black is winning here. If I had an engine, I would teach it that the evaluation at the root is +3 full pawns in black's favour.

[Uri Blass]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

engines are better than humans so I trust stockfish here.

I also think that there was a mistake of white inside the 35 plies when SF evaluated white as better so it is not a case when SF need to search additional 35 plies to see the truth.

I also do not believe that the gap between stockfish (after hours of search) and the ultimate truth is at least 500 elo

There are positions when engines are blind even after hours of search but they do not happen often in games.

For the first time, I do not agree with a single thing you say.

1. Humans are in general better than engines, engines might have some tactical advantage, but humans are much better positionally.

2. It actually might need more than 70 plies here, as there is, at worst, a very very long series of queen checks, which prevents SF and other engines to make any reasonable sense of what goes on. For the time checks are given (maybe some 40-50 plies in a couple of series of checks) SF counts only the material advantage, does not see black promotion, so it sticks to its score. However, I have 2 reasons to believe black wins here:

- central passers are much faster in advancing than edge passers
- the pieces of lower power easily control squares both in front of the own and in front of the enemy passers, which is not true for the queen

3. I think the gap might be even bigger.

4. Positions where engines are blind even after hours of search do not happen often in games, simply because the engines know only a single style of play, an open one. And opening books are also devised accordingly. Really, I almost do not see closed positions in engine games, less than 0.1% for sure. However, when I play against SF, 95% of my games are closed. If I want to close the game, I simply close it and SF can not prevent this. So simply positions engines do not understand do not happen often in tests, because of the engine inability to play differently. But it is not only closed positions, there are at least 15 other totally different playing styles that have never happened in engine games and that engines will not understand even after a very long search.

I disagree

Engines play also good positional moves thanks to deep search.
Humans can better explain why some move is strong
but if we talk about evaluation I trust more engines.

There are cases that I can say that the position is a draw inspite of big advantage by evaluation because the engines do not understand fortress but almost never cases when the position is won for black and engines say won for white.

If the engine is wrong about the winner then I at least expect to see that the score for white is going down and not going up and I do not see it in this case.

[Lyudmil Tsvetkov]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

engines are better than humans so I trust stockfish here.

I also think that there was a mistake of white inside the 35 plies when SF evaluated white as better so it is not a case when SF need to search additional 35 plies to see the truth.

I also do not believe that the gap between stockfish (after hours of search) and the ultimate truth is at least 500 elo

There are positions when engines are blind even after hours of search but they do not happen often in games.

For the first time, I do not agree with a single thing you say.

1. Humans are in general better than engines, engines might have some tactical advantage, but humans are much better positionally.

2. It actually might need more than 70 plies here, as there is, at worst, a very very long series of queen checks, which prevents SF and other engines to make any reasonable sense of what goes on. For the time checks are given (maybe some 40-50 plies in a couple of series of checks) SF counts only the material advantage, does not see black promotion, so it sticks to its score. However, I have 2 reasons to believe black wins here:

- central passers are much faster in advancing than edge passers
- the pieces of lower power easily control squares both in front of the own and in front of the enemy passers, which is not true for the queen

3. I think the gap might be even bigger.

4. Positions where engines are blind even after hours of search do not happen often in games, simply because the engines know only a single style of play, an open one. And opening books are also devised accordingly. Really, I almost do not see closed positions in engine games, less than 0.1% for sure. However, when I play against SF, 95% of my games are closed. If I want to close the game, I simply close it and SF can not prevent this. So simply positions engines do not understand do not happen often in tests, because of the engine inability to play differently. But it is not only closed positions, there are at least 15 other totally different playing styles that have never happened in engine games and that engines will not understand even after a very long search.

I disagree

Engines play also good positional moves thanks to deep search.
Humans can better explain why some move is strong
but if we talk about evaluation I trust more engines.

There are cases that I can say that the position is a draw inspite of big advantage by evaluation because the engines do not understand fortress but almost never cases when the position is won for black and engines say won for white.

If the engine is wrong about the winner then I at least expect to see that the score for white is going down and not going up and I do not see it in this case.

Actually, from what I see from Louis output, SF evaluation jumps up and down as the search progresses. At depth 48 it is 275cps, at depth 49 313cps, at depth 50 it goes back to 275cps, only to reach 350cps at depth 54. Who knows what happens at depth 56?
Besides, in the lines shown, SF changes its mind about the best moves at plies 2,3 and 4, how can I trust then its assessment for bigger plies?

[Uri Blass]

looking at the game I never saw more than +2 for white when in the analysis I saw even scores near +3 pawns for white and this is the reason that I believe white is winning.

Looking at the game I will try to find the losing mistake for white
Note that black's moves also were not optimal so practically the lines that stockfish needs to search are shorter.

first analysis suggests that

61...Ba3 was winning for black instead of 61...Rd5+ that is slower

I also can say that after 58...d2 black had at least a draw and probably the mistake of white was earlier.

Note that old stockfish did a mistake by giving a bonus for the a4 b4 pawns and the last change is not to give a bonus for protected passed pawns or 2 connected passed pawns in the same file.

I will check if 61.Kf3 could save the game or no.

[Uri Blass]

Hi Oliver.

I would like to analyse, but I do not have the time and resources. It is very demanding.

Absolutely, chess is deep and beautiful, and modern engines are only halfway going to the ultimate truth, maybe some 500-1000 elo, even more away.

It is an illusion to trust modern engines in all situations, there are multitude of position types top engines fully fail to understand.

I do not trust SF in the above position, because in the actual game some 35 plies on SF still thought white is winning by a large margin; and when you add another 35 plies SF computed in the game, this means that, in order to see a clear outcome, SF will need more than 70 plies! And quite possibly those plies will be infested with bugs. I am stubborn, I still say black is winning this.

engines are better than humans so I trust stockfish here.

I also think that there was a mistake of white inside the 35 plies when SF evaluated white as better so it is not a case when SF need to search additional 35 plies to see the truth.

I also do not believe that the gap between stockfish (after hours of search) and the ultimate truth is at least 500 elo

There are positions when engines are blind even after hours of search but they do not happen often in games.

For the first time, I do not agree with a single thing you say.

1. Humans are in general better than engines, engines might have some tactical advantage, but humans are much better positionally.

2. It actually might need more than 70 plies here, as there is, at worst, a very very long series of queen checks, which prevents SF and other engines to make any reasonable sense of what goes on. For the time checks are given (maybe some 40-50 plies in a couple of series of checks) SF counts only the material advantage, does not see black promotion, so it sticks to its score. However, I have 2 reasons to believe black wins here:

- central passers are much faster in advancing than edge passers
- the pieces of lower power easily control squares both in front of the own and in front of the enemy passers, which is not true for the queen

3. I think the gap might be even bigger.

4. Positions where engines are blind even after hours of search do not happen often in games, simply because the engines know only a single style of play, an open one. And opening books are also devised accordingly. Really, I almost do not see closed positions in engine games, less than 0.1% for sure. However, when I play against SF, 95% of my games are closed. If I want to close the game, I simply close it and SF can not prevent this. So simply positions engines do not understand do not happen often in tests, because of the engine inability to play differently. But it is not only closed positions, there are at least 15 other totally different playing styles that have never happened in engine games and that engines will not understand even after a very long search.

I disagree

Engines play also good positional moves thanks to deep search.
Humans can better explain why some move is strong
but if we talk about evaluation I trust more engines.

There are cases that I can say that the position is a draw inspite of big advantage by evaluation because the engines do not understand fortress but almost never cases when the position is won for black and engines say won for white.

If the engine is wrong about the winner then I at least expect to see that the score for white is going down and not going up and I do not see it in this case.

Actually, from what I see from Louis output, SF evaluation jumps up and down as the search progresses. At depth 48 it is 275cps, at depth 49 313cps, at depth 50 it goes back to 275cps, only to reach 350cps at depth 54. Who knows what happens at depth 56?
Besides, in the lines shown, SF changes its mind about the best moves at plies 2,3 and 4, how can I trust then its assessment for bigger plies?

There is up and down but the general tendency is to go up and this cause me to believe that at least white is not losing the game(I am less sure of winning the game but 3.50 pawns advantage stongly suggest winning the game unlike the 1.xx evaluations during the game).

[Uri Blass]

Note that during the game white clearly has the option to repeat the position but white did not want a draw so he did not try to do it.

I believe that the
The queen can get at least perpetual check.

So far latest stockfish does not see a win for black after 61.Kf3 and suggest at depth 50 the repetition move 61...Rd5