It is nice! Less ELO with threads for new network architecture.

https://github.com/official-stockfish/Stockfish/issues/3727
http://talkchess.com/forum3/viewtopic.php?f=2&t=78539&start=10

It is nice to see my results are being recognized after all this time. Since the very release of NNUE. So I guess it will still take the chess programming community time to understand this is not a issue with NNUE being close to perfect play.  

As if we were close to perfect play as in some of the comments suggest as a reason why. The best test would be being able to solve most chess problems, or simple endgame positions where we know what perfect play looks like with seven man table bases.  But since we are not, and would not be a issue. if chess engines were close to perfect play.

This is just a issue with NNUE, and scaling. As we are far from perfect play. And I will leave the comment at that.

Sorry I am away on vacation for a time. So testing will resume when I am back.

It would be possible to be "unbeatable" without "perfect play".

TheSelfImprover wrote:

It would be possible to be "unbeatable" without "perfect play".

What you wrote is a contradiction in logic.

No! And we do not know, and can never know. As chess is not solvable.

This is very simple in concept.

1. If chess is a draw with perfect play. Then it would take a perfect player to always draw against perfect play. Otherwise it would not be perfect play. And engines can not do this with the perfect play we have today. 7 man table bases. Chess engines fail this test even in very simple endgames against perfect play.

2. If chess is a win for either white or black. Then against perfect play the best you can hope to score is 50% in a game pair. But always losing the losing side.

3. And we know for a fact that engines do not play perfect chess. As even the best engines tell us so every game by their evaluation. Every time a engine gives a incorrect evaluation in a position. We know the engines is only guessing, and giving an approximation. And that is far from perfect play.

4. And All the best engines are Shannon type B engines. Meaning given even a trillion years per move on todays fastest computers. It still would be far from perfect play. As chess is a 100% tactical game. And Shannon type B engines prune 99.999999999999999999999999999999999999999999999999999999999999999999.... percent of all moves. Unless you really think chess engines are doing a real 30 ply search in just a matter of seconds. Like on my system.  

mwyoung wrote:

TheSelfImprover wrote:

It would be possible to be "unbeatable" without "perfect play".

What you wrote is a contradiction in logic.

No! And we do not know, and can never know. As chess is not solvable.

This is very simple in concept.

1. If chess is a draw with perfect play. Then it would take a perfect player to always draw against perfect play. Otherwise it would not be perfect play. And engines can not do this with the perfect play we have today. 7 man table bases. Chess engines fail this test even in very simple endgames against perfect play.

2. If chess is a win for either white or black. Then against perfect play the best you can hope to score is 50% in a game pair. But always losing the losing side.

3. And we know for a fact that engines do not play perfect chess. As even the best engines tell us so every game by their evaluation. Every time a engine gives a incorrect evaluation in a position. We know the engines is only guessing, and giving an approximation. And that is far from perfect play.

4. And All the best engines are Shannon type B engines. Meaning given even a trillion years per move on todays fastest computers. It still would be far from perfect play. As chess is a 100% tactical game. And Shannon type B engines prune 99.999999999999999999999999999999999999999999999999999999999999999999.... percent of all moves. Unless you really think chess engines are doing a real 30 ply search in just a matter of seconds. Like on my system.  

1)The fact that we do not know does not prove that chess engines are not unbeatable.
The only proof that they are not unbeatable can be simply by beating them.

2)There are positions from 7 man table bases when chess engines do not play perfectly but it is possible that you cannot get one of these position in a game against them so it does not prove that chess engines are not unbeatable.

3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake) but they are not unbeatable because it is possible to play different moves that cause them to do a mistake.
My guess is that this is what happens and not that stockfish or dragon at 120/40 time control are unbeatable but I have no proof for it.
The only way to prove that it is possible to beat Stockfish or Dragon at 120/40 is simply to beat them.

I would like to see some engine that can beat stockfish by playing against it and learning but I do not see this type of testing.
The idea is simple.
You play game stockfish against Dragon(let say with no opening book and get a draw.
You remember the best score that Stockfish got in the game and you remember the moves that dragon played in the game and add the moves that dragon played to the opening book.
In the next game you try to play some different move relative to the best game that you have that is initially the first game.
If the best score that you get is higher than the best score that you got in the best game than the new game become the best game.
Otherwise you do not replace the best game and try to change a different move in the best game hoping to get a new best game.

You finish the process when you get a win or when every change that you make to one of the moves in the best game does not improve the score.
I wonder if this learning is enough to produce a win or maybe you need some better learning and changing a single move and playing the rest Stockfish-Dragon is not enough.

Uri Blass wrote:

mwyoung wrote:

TheSelfImprover wrote:

It would be possible to be "unbeatable" without "perfect play".

What you wrote is a contradiction in logic.

No! And we do not know, and can never know. As chess is not solvable.

This is very simple in concept.

1. If chess is a draw with perfect play. Then it would take a perfect player to always draw against perfect play. Otherwise it would not be perfect play. And engines can not do this with the perfect play we have today. 7 man table bases. Chess engines fail this test even in very simple endgames against perfect play.

2. If chess is a win for either white or black. Then against perfect play the best you can hope to score is 50% in a game pair. But always losing the losing side.

3. And we know for a fact that engines do not play perfect chess. As even the best engines tell us so every game by their evaluation. Every time a engine gives a incorrect evaluation in a position. We know the engines is only guessing, and giving an approximation. And that is far from perfect play.

4. And All the best engines are Shannon type B engines. Meaning given even a trillion years per move on todays fastest computers. It still would be far from perfect play. As chess is a 100% tactical game. And Shannon type B engines prune 99.999999999999999999999999999999999999999999999999999999999999999999.... percent of all moves. Unless you really think chess engines are doing a real 30 ply search in just a matter of seconds. Like on my system.  

1)The fact that we do not know does not prove that chess engines are not unbeatable.
The only proof that they are not unbeatable can be simply by beating them.

2)There are positions from 7 man table bases when chess engines do not play perfectly but it is possible that you cannot get one of these position in a game against them so it does not prove that chess engines are not unbeatable.

3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake) but they are not unbeatable because it is possible to play different moves that cause them to do a mistake.
My guess is that this is what happens and not that stockfish or dragon at 120/40 time control are unbeatable but I have  no proof for it.
The only way to prove that it is possible to beat Stockfish or Dragon at 120/40 is simply to beat them.

I would like to see some engine that can beat stockfish by playing against it and learning but I do not see this type of testing.
The idea is simple.
You play game stockfish against Dragon(let say with no opening book and get a draw.
You remember the best score that Stockfish got in the game and you remember the moves that dragon played in the game and add the moves that dragon played to the opening book.
In the next game you try to play some different move relative to the best game that you have that is initially the first game.
If the best score that you get is higher than the best score that you got in the best game than the new game become the best game.
Otherwise you do not replace the best game and try to change a different move in the best game hoping to get a new best game.

You finish the process when you get a win or when every change that you make to one of the moves in the best game does not improve the score.
I wonder if this learning is enough to produce a win or maybe you need some better learning and changing a single move and playing the rest Stockfish-Dragon is not enough.

What you wrote is ridiculous.

"3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake)

I will just take one.

Then by your logic chess draws = perfect play.

This logic is clearly false! As it only means the other player was not able to understand the mistake and refuted it.

This is like me saying I am unbeatable and play perfect! if I only play my dog chess.

mwyoung wrote:

Uri Blass wrote:

mwyoung wrote:

TheSelfImprover wrote:

It would be possible to be "unbeatable" without "perfect play".

What you wrote is a contradiction in logic.

No! And we do not know, and can never know. As chess is not solvable.

This is very simple in concept.

1. If chess is a draw with perfect play. Then it would take a perfect player to always draw against perfect play. Otherwise it would not be perfect play. And engines can not do this with the perfect play we have today. 7 man table bases. Chess engines fail this test even in very simple endgames against perfect play.

2. If chess is a win for either white or black. Then against perfect play the best you can hope to score is 50% in a game pair. But always losing the losing side.

3. And we know for a fact that engines do not play perfect chess. As even the best engines tell us so every game by their evaluation. Every time a engine gives a incorrect evaluation in a position. We know the engines is only guessing, and giving an approximation. And that is far from perfect play.

4. And All the best engines are Shannon type B engines. Meaning given even a trillion years per move on todays fastest computers. It still would be far from perfect play. As chess is a 100% tactical game. And Shannon type B engines prune 99.999999999999999999999999999999999999999999999999999999999999999999.... percent of all moves. Unless you really think chess engines are doing a real 30 ply search in just a matter of seconds. Like on my system.  

1)The fact that we do not know does not prove that chess engines are not unbeatable.
The only proof that they are not unbeatable can be simply by beating them.

2)There are positions from 7 man table bases when chess engines do not play perfectly but it is possible that you cannot get one of these position in a game against them so it does not prove that chess engines are not unbeatable.

3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake) but they are not unbeatable because it is possible to play different moves that cause them to do a mistake.
My guess is that this is what happens and not that stockfish or dragon at 120/40 time control are unbeatable but I have  no proof for it.
The only way to prove that it is possible to beat Stockfish or Dragon at 120/40 is simply to beat them.

I would like to see some engine that can beat stockfish by playing against it and learning but I do not see this type of testing.
The idea is simple.
You play game stockfish against Dragon(let say with no opening book and get a draw.
You remember the best score that Stockfish got in the game and you remember the moves that dragon played in the game and add the moves that dragon played to the opening book.
In the next game you try to play some different move relative to the best game that you have that is initially the first game.
If the best score that you get is higher than the best score that you got in the best game than the new game become the best game.
Otherwise you do not replace the best game and try to change a different move in the best game hoping to get a new best game.

You finish the process when you get a win or when every change that you make to one of the moves in the best game does not improve the score.
I wonder if this learning is enough to produce a win or maybe you need some better learning and changing a single move and playing the rest Stockfish-Dragon is not enough.

What you wrote is ridiculous.

"3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake)

I will just take one.

Then by your logic chess draws = perfect play.  

This logic is clearly false! As it only means the other player was not able to understand the mistake and refuted it.

This is like me saying I am unbeatable and play perfect! if I only play my dog chess.  

No
I do not say draws=perfect play
draws can be perfect play and can be non perfect play.

I only say that it is possible that the draws are perfect play in the meaning of not doing mistakes.
It is possible does not mean that I am sure that it is the case but only that I am not sure that it is not the case.

It is also possible that the draws happen when one side make a losing mistake and the second side does not take advantage of the mistake.
My guess is that many of the draws are perfect play but it is not something that I am sure about it.

Note that perfect play does not mean that the engines are unbeatable because engine A may play perfect moves
that do not cause engine B to make a losing mistake when it is possible to play different perfect moves that cause engine B to make a losing mistake.

Uri Blass wrote:

mwyoung wrote:

Uri Blass wrote:

mwyoung wrote:

TheSelfImprover wrote:

It would be possible to be "unbeatable" without "perfect play".

What you wrote is a contradiction in logic.

No! And we do not know, and can never know. As chess is not solvable.

This is very simple in concept.

1. If chess is a draw with perfect play. Then it would take a perfect player to always draw against perfect play. Otherwise it would not be perfect play. And engines can not do this with the perfect play we have today. 7 man table bases. Chess engines fail this test even in very simple endgames against perfect play.

2. If chess is a win for either white or black. Then against perfect play the best you can hope to score is 50% in a game pair. But always losing the losing side.

3. And we know for a fact that engines do not play perfect chess. As even the best engines tell us so every game by their evaluation. Every time a engine gives a incorrect evaluation in a position. We know the engines is only guessing, and giving an approximation. And that is far from perfect play.

4. And All the best engines are Shannon type B engines. Meaning given even a trillion years per move on todays fastest computers. It still would be far from perfect play. As chess is a 100% tactical game. And Shannon type B engines prune 99.999999999999999999999999999999999999999999999999999999999999999999.... percent of all moves. Unless you really think chess engines are doing a real 30 ply search in just a matter of seconds. Like on my system.  

1)The fact that we do not know does not prove that chess engines are not unbeatable.
The only proof that they are not unbeatable can be simply by beating them.

2)There are positions from 7 man table bases when chess engines do not play perfectly but it is possible that you cannot get one of these position in a game against them so it does not prove that chess engines are not unbeatable.

3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake) but they are not unbeatable because it is possible to play different moves that cause them to do a mistake.
My guess is that this is what happens and not that stockfish or dragon at 120/40 time control are unbeatable but I have  no proof for it.
The only way to prove that it is possible to beat Stockfish or Dragon at 120/40 is simply to beat them.

I would like to see some engine that can beat stockfish by playing against it and learning but I do not see this type of testing.
The idea is simple.
You play game stockfish against Dragon(let say with no opening book and get a draw.
You remember the best score that Stockfish got in the game and you remember the moves that dragon played in the game and add the moves that dragon played to the opening book.
In the next game you try to play some different move relative to the best game that you have that is initially the first game.
If the best score that you get is higher than the best score that you got in the best game than the new game become the best game.
Otherwise you do not replace the best game and try to change a different move in the best game hoping to get a new best game.

You finish the process when you get a win or when every change that you make to one of the moves in the best game does not improve the score.
I wonder if this learning is enough to produce a win or maybe you need some better learning and changing a single move and playing the rest Stockfish-Dragon is not enough.

What you wrote is ridiculous.

"3)Assuming chess is a draw with perfect play then it is possible that chess engines play perfect in the draws that you have in stokfish-dragon match (when perfect means they do not do a losing mistake)

I will just take one.

Then by your logic chess draws = perfect play.  

This logic is clearly false! As it only means the other player was not able to understand the mistake and refuted it.

This is like me saying I am unbeatable and play perfect! if I only play my dog chess.  

No
I do not say draws=perfect play
draws can be perfect play and can be non perfect play.

I only say that it is possible that the draws are perfect play in the meaning of not doing mistakes.
It is possible does not mean that I am sure that it is the case but only that I am not sure that it is not the case.

It is also possible that the draws happen when one side make a losing mistake and the second side does not take advantage of the mistake.
My guess is that many of the draws are perfect play but it is not something that I am sure about it.

Note that perfect play does not mean that the engines are unbeatable because engine A may play perfect moves
that do not cause engine B to make a losing mistake when it is possible to play different perfect moves that cause engine B to make a losing mistake.

"I only say that it is possible that the draws are perfect play in the meaning of not doing mistakes."

How would you know!. And it is meaningless. if chess is a forced win with perfect play. As that is also possible. And that would say the engines are making mistakes with every draw.

Here is what can be proven. No chess engine is able to pass the test of playing 7 man or less chess perfectly. So why would anyone think that chess engines can play perfectly in every game with more then 7 men. As that would be far more complex then with 7 man or less.

And why would you think Stockfish. A type B Shannon engine could every play perfect chess. Do you know that the type B design of a chess engine. Was never designed to ever play perfect chess. But to only beat humans.

So you could never prove perfect play by having 2 imperfect engines playing each other. No matter the draw ratio. As we know they are imperfect players. As can be proven.