Results of the top ten strongest engines tourney at CCRL

mwyoung wrote:

Admin wrote:

mwyoung wrote:

For the purpose of just lowing the draw rate and distorting the game outcome data.

So what's your view on TCEC, the real world championship or not?

I have wrote about this 100 times. TCEC is not watched, or paid attention too. I could not tell you even who is playing.

TCEC is TRASH for using gamed openings.

And I know who the world champ is long before TCEC existed, and I know today.

Do you realize your view is a minority one?

mwyoung wrote:

Admin wrote:

mwyoung wrote:

Not ONE garbage opening was played!

This was 1998 just one year after the Deep Blue victory over Kasparov, computer chess still in the stone age. The goal was to rely on the then strongest point of a computer which at the same time is the weakest part of the human, create tactics on the board, not necessarily by playing the strongest move. Whether that worked or not is up to everybody's opinion, point is it did not work at all against other engines. It was a strategy that worked IMO.

Fast forward 25 years. it's 2023, Rebel is 800 elo stronger, the result: draws, draws and draws, boring....

So now I am busy to create a net that plays more risky at the cost of some (not much) elo in the hope to see more fireworks on the board and to reduce draws and if the inevitable draws happen they are at least less boring. It's another strategy, this time not to win but to keep my interest in computer chess alive.

Look, I am not arguing the standard way of testing with balanced openings, that is just fine and should be kept. I am advocating a secondary important rating list to keep computer chess alive. None of the 100 positions I used for the GRL is lost and look at the low draw rate and replay some games.

You are in the wrong line of work as a engine tester.

Do you actually read what I have said?

Quote :

Draws are part of the game. And there is nothing wrong with draws..... And what do you expect, the engines are all around the same Elo level, because they use almost the exact same evaluations. So at this point we are evaluating the search efficiency of the chess engine.

You have no idea about net (evaluation) building.

Quote :

THEN YOU ARE SHOCKED at the draw rate.

Wrong again.

High draw rates are a natural effect of playing strength.

Why do you think I am shocked?

Please refrain from putting words in my mouth.

Thank you.

mwyoung wrote:

Admin wrote:

mwyoung wrote:

Not ONE garbage opening was played!

This was 1998 just one year after the Deep Blue victory over Kasparov, computer chess still in the stone age. The goal was to rely on the then strongest point of a computer which at the same time is the weakest part of the human, create tactics on the board, not necessarily by playing the strongest move. Whether that worked or not is up to everybody's opinion, point is it did not work at all against other engines. It was a strategy that worked IMO.

Fast forward 25 years. it's 2023, Rebel is 800 elo stronger, the result: draws, draws and draws, boring....

So now I am busy to create a net that plays more risky at the cost of some (not much) elo in the hope to see more fireworks on the board and to reduce draws and if the inevitable draws happen they are at least less boring. It's another strategy, this time not to win but to keep my interest in computer chess alive.

Look, I am not arguing the standard way of testing with balanced openings, that is just fine and should be kept. I am advocating a secondary important rating list to keep computer chess alive. None of the 100 positions I used for the GRL is lost and look at the low draw rate and replay some games.

You are in the wrong line of work as a engine tester. Draws are part of the game. And there is nothing wrong with draws..... And what do you expect, the engines are all around the same Elo level, because they use almost the exact same evaluations. So at this point we are evaluating the search efficiency of the chess engine.

THEN YOU ARE SHOCKED at the draw rate.

The draw rate is also going to be high if you play the same engine against itself and one engine get hardware that is 10 times faster.
It was not the case some years ago so I suspect it is high because we are close to have some unbeateable engine in chess(there are of course positions when the engines do not find the right move but playing perfectly may not be enough to get them and you may need some smart to use some weakness of the engines to get them).

I think some competition of beating stockfish with fixed nodes per move and no book and fixed hash of 4096 mbytes(so stockfish is a determinstic engine) may be more interesting to see if we get closer to having some unbeatable chess engines.

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

Admin wrote:

mwyoung wrote:

Admin wrote:

mwyoung wrote:

For the purpose of just lowing the draw rate and distorting the game outcome data.

So what's your view on TCEC, the real world championship or not?

I have wrote about this 100 times. TCEC is not watched, or paid attention too. I could not tell you even who is playing.

TCEC is TRASH for using gamed openings.

And I know who the world champ is long before TCEC existed, and I know today.

Do you realize your view is a minority one?

And amount those who has my knowledge and experience in testing all the chess computers and chess engines since 1980....

Almost NONE!

These are the same people that think wrongly.

A high draw rate equals chess being solved by chess computers.

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.
Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Ed that is 2 people that really understand the issues causing the draw rate. And there are many more like us. And the solution is not in using gamed openings. The testing data is telling us the truth.

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Admin wrote:

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

You know that is the hill you are going to die on Ed. When I get home I will prove it with game data!

The mistake is a word that makes no sense.
A mistake is always within a range of moves = within a horizon.
A mistake in a range of 10 plies can be a win in a range of 15 plies.


Then the mistake is called sacrifice.

So what is a mistake ? The only mistake that is NOT in a range of anything is a forced mate.
Here you can say the move was a mistake because it leads to mate in 5.

But outside this range, mistake is a word making no sense.


Is 1.e3 a mistake ?
Or 1.h3 ?!

Who knows ?!
Stockfish does not know either.



IMO CSTals idea was to play interesting.
I would therefore play unbalanced openings and gambit openings.
Yes that will drop down playing strength overall.
But i would give more fun games.

Chris Whittington wrote:

Admin wrote:

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.





I’m with Ed on this. Many engines are similar but that’s not the reason for the draw rate. SF and LC0 are about as dissimilar as can be, but they draw mostly against each other. Komodo has some imbalance knowledge built into it, so it is not exactly the same as SF, but it also has a high draw rate. Computer chess winning is about getting a small advantage, increasing it to a tipping point where it’s then a win. Deeper searches (with a decent eval) usually are able to find moves that avoid any tipping point until the game runs out due to material trades.

Chess is not an unstable game (unlike, say, Shogi). Small advantages do not exponentially run into large ones because material trades down as the game progresses

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Because you are over looking one important fact. Simular alsio works with engine ratings. Not just the same engine.

But for consistency we will prove it with both. But I have already tested this both ways.

Because some here do not just run with blind assumptions.

We try to prove this or disprove it. And sees what the data shows.

And what is shocking is this data has been seen for years in both high level human chess., and high level computer chess.

This is no revelation.

Mclane wrote:

The mistake is a word that makes no sense.
A mistake is always within a range of moves = within a horizon.
A mistake in a range of 10 plies can be a win in a range of 15 plies.


Then the mistake is called sacrifice.

So what is a mistake ? The only mistake that is NOT in a range of anything is a forced mate.
Here you can say the move was a mistake because it leads to mate in 5.

But outside this range, mistake is a word making no sense.


Is 1.e3 a mistake ?
Or 1.h3 ?!

Who knows ?!
Stockfish does not know either.

I won't play judge when it is about your thoughts about computerchess, you being a renowned expert on this topic.

When it is about chess itself though, you could profit very much from watching some freely availlable stuff on what it is about. Someone who is very good at explaining ideas in an accessable way is Yasser Seirawan.

Here is a nice video on an opening variation, usually a pretty boring topic for people. If you invest 40 minutes, you'll know more about the game, promised.

https://www.youtube.com/watch?v=TvipCjxqGiM

Yasser seirawan talks about ideas. But do chess engines have ideas ?!

I thought CSTal was about playing interesting. Not about playing the strongest.

Its IMO almost unimportant if it has a high elo or not, as long as it plays spectacular chess.

Of course doing that against the top 5 is difficult.
Against the top 10 it is possible.
And against the 11-20 it should be possible.

Against human beeings even more.

It seems to me that the question here is: are we bumping into the upper limits of chess, or is it possible to build a vastly superior chess computer that would win most of its games with today's top engines?

I am strongly on the "upper limits" side:

* after nearly 70 years of continual advancement in computer chess, no way has been found to win material from the opening position

* lesser games are now completely solved (Connect 4, draughts (checkers)) etc. In each case, it became clear which way the outcome was going long before the final solution arrived

* there has long been a correlation between the elo rating of equal players and the draw rate - and it's never stopped moving towards 1

* consequently, we have a good idea what the upper limit of the elo rating system is

* no game has ever sprung a surprise and produced a different result from the one expected by analysing results from top humans

mwyoung is correct on a fundamentally important point: when engines are not permitted to play the openings of their choice, they're no longer playing classic chess.

When i watch eng-eng chess i am shocked how fast they have tablebase access even when in late opening stage.
So no wonder they play draw. If both use tablebases and both play „best moves“ without risking material for a plan, the game ends in draw.

mwyoung wrote:

And what is shocking is this data has been seen for years in both high level human chess., and high level computer chess.

This is no revelation.

Well then, what's the problem?

The higher the elo pool, the higher the draw rate, true for human chess and true for computer chess. Doesn't mean engines are similar. Have you followed Graham's tournament Rebel playing Chess System Tal? Did you the noticed the big score differences between both engines? And notable with a similar search?

NNUE programming is a different animal, even us the programmers don't understand what happens inside, too complex for the human mind, we only know that it works, improvements come from experiments, it's like looking for treasures in a dark cave with a flashlight. And even what worked for HCE to measure eval similarity doesn't work any longer for NNUE.

You can't have proof.

Chris Whittington wrote:

Admin wrote:

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.





I’m with Ed on this. Many engines are similar but that’s not the reason for the draw rate. SF and LC0 are about as dissimilar as can be, but they draw mostly against each other. Komodo has some imbalance knowledge built into it, so it is not exactly the same as SF, but it also has a high draw rate. Computer chess winning is about getting a small advantage, increasing it to a tipping point where it’s then a win. Deeper searches (with a decent eval) usually are able to find moves that avoid any tipping point until the game runs out due to material trades.

Chess is not an unstable game (unlike, say, Shogi). Small advantages do not exponentially run into large ones because material trades down as the game progresses

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Let us disprove this nonsense.

First we will address this fact I posted, and was disputed about TCEC.

From TCEC

Q:  To what degree do you bias openings?

A: In each season’s Superfinal Jeroen has a free hand to do whatever he thinks will result in an interesting and varied contest.  Sometimes his book-exit evaluations will exceed +/-1.00; sometimes he will offer speculative gambits.  His goal is to keep the draw-rate in the 65-80% range each season without a surfeit of one-sided openings.

And even TCEC agrees with our position. Draws are caused by closely match chess engines.

"if very strong, closely matched programs are playing at long time controls with no bias that invites a draw-rate approaching 90%."

I am not sure were the myth started that draws somehow indicated near flawless play by the chess engine. The fact is this has never been true. Draws happen because a chess engine with the same rating. Does not know how to exploit the errors of a similar chess engine in rating and or functions.

Now we will directly disprove this statement.
 
This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.[/quote]

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out. --- Ed.


Chess engine A plays Chess engine A duplicate. The draw rate scored is 93%

Now was the draw rate caused by the chess engine making less and less mistakes. Or because the chess engine does not know how to exploit the errors of the other chess engine?  

And how can we prove this to see who is correct.

Simple. Have chess engine A play a stronger chess engine. If the draws were caused by mistake free chess. The draws should remain the same.

Lets see what happens....

Chess engine A now plays Chess engine B. Draw rate is now 57%. And chess engine A is crushed.

We will end this with a Mclane quote describing this wrong conclusion reached by circular logic.

"People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other."

This demonstration was played with Perfect Book 2021 to six moves. And played in gauntlet mode.

Code:

Result:
--------------------------------------------------------------------------------------
  #  name                      games            wins           draws          losses           score    elo    +    -
  1. Stockfish 16                200       85 100.0%      114 100.0%        1 100.0%       142 71.0%     78   26   25
  2. Stockfish 11 64 BMI2 dup    200       11 100.0%      186 100.0%        3 100.0%       104 52.0%    -34   22   22
  3. Stockfish 11 64 BMI2        400        4 100.0%      300 100.0%       96 100.0%       154 38.5%    -44   17   17

Cross table:
--------------------------------------------------------------------------------------
  #  name                                 score   games         1         2         3
  1. Stockfish 16                     142 71.0%     200         x               142.0
  2. Stockfish 11 64 BMI2 dup         104 52.0%     200                   x     104.0
  3. Stockfish 11 64 BMI2             154 38.5%     400      58.0      96.0         x

Tech:
--------------------------------------------------------------------------------------

Tech (average nodes, depths, time/m per move, others per game), counted for computing moves only, ignored moves with zero nodes:
  #  name                        nodes/m         NPS  depth/m   time/m    moves     time
  1. Stockfish 16                 17892K    10242490     40.6      1.7     69.2    120.8
  2. Stockfish 11 64 BMI2 dup     26977K    15238064     34.4      1.8     52.8     93.5
  3. Stockfish 11 64 BMI2         27141K    15509692     33.7      1.7     60.9    106.6
     all ---                      23908K    13961837     35.8      1.8     61.0    106.9
 

Mclane wrote:

The mistake is a word that makes no sense.
A mistake is always within a range of moves = within a horizon.
A mistake in a range of 10 plies can be a win in a range of 15 plies.


Then the mistake is called sacrifice.

So what is a mistake ? The only mistake that is NOT in a range of anything is a forced mate.
Here you can say the move was a mistake because it leads to mate in 5.

But outside this range, mistake is a word making no sense.


Is 1.e3 a mistake ?
Or 1.h3 ?!

Who knows ?!
Stockfish does not know either.



IMO CSTals idea was to play interesting.
I would therefore play unbalanced openings and gambit openings.
Yes that will drop down playing strength overall.
But i would give more fun games.

A mistake is a move that change the result of the game with perfect play.
I am not sure but I believe 1.e3 or 1.h3 are not mistakes by this definition.

Usually positions that stockfish evaluate as less than 0.5 pawns are draws and if the result remains less than 0.5 pawns then the result is still a draw

For a better investigation of the subject we need a tool that get some input and generate output so we have some estimate for probabilities of win and draw and loss for every evaluation based on the chess engine that we choose and the time control that we choose and games of the engine against itself.

Of course if we choose a long time control the process may take too much time to get meaningfull results.


Input can be:
1)games in pgn(I want the number of them to be big but not too big because if it is too big it is going to take many years to generate the output)
2)some engine
3)some time control to run engine-engine games from all positions

mwyoung wrote:

Chris Whittington wrote:

Admin wrote:

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.





I’m with Ed on this. Many engines are similar but that’s not the reason for the draw rate. SF and LC0 are about as dissimilar as can be, but they draw mostly against each other. Komodo has some imbalance knowledge built into it, so it is not exactly the same as SF, but it also has a high draw rate. Computer chess winning is about getting a small advantage, increasing it to a tipping point where it’s then a win. Deeper searches (with a decent eval) usually are able to find moves that avoid any tipping point until the game runs out due to material trades.

Chess is not an unstable game (unlike, say, Shogi). Small advantages do not exponentially run into large ones because material trades down as the game progresses

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Let us disprove this nonsense.

First we will address this fact I posted, and was disputed about TCEC.

From TCEC

Q:  To what degree do you bias openings?

A: In each season’s Superfinal Jeroen has a free hand to do whatever he thinks will result in an interesting and varied contest.  Sometimes his book-exit evaluations will exceed +/-1.00; sometimes he will offer speculative gambits.  His goal is to keep the draw-rate in the 65-80% range each season without a surfeit of one-sided openings.

And even TCEC agrees with our position. Draws are caused by closely match chess engines.

"if very strong, closely matched programs are playing at long time controls with no bias that invites a draw-rate approaching 90%."

I am not sure were the myth started that draws somehow indicated near flawless play by the chess engine. The fact is this has never been true. Draws happen because a chess engine with the same rating. Does not know how to exploit the errors of a similar chess engine in rating and or functions.

Now we will directly disprove this statement.
 
This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out. --- Ed.


Chess engine A plays Chess engine A duplicate. The draw rate scored is 93%

Now was the draw rate caused by the chess engine making less and less mistakes. Or because the chess engine does not know how to exploit the errors of the other chess engine?  

And how can we prove this to see who is correct.

Simple. Have chess engine A play a stronger chess engine. If the draws were caused by mistake free chess. The draws should remain the same.

Lets see what happens....

Chess engine A now plays Chess engine B. Draw rate is now 57%. And chess engine A is crushed.

We will end this with a Mclane quote describing this wrong conclusion reached by circular logic.

"People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other."

This demonstration was played with Perfect Book 2021 to six moves. And played in gauntlet mode.

Code:

Result:
--------------------------------------------------------------------------------------
  #  name                      games            wins           draws          losses           score    elo    +    -
  1. Stockfish 16                200       85 100.0%      114 100.0%        1 100.0%       142 71.0%     78   26   25
  2. Stockfish 11 64 BMI2 dup    200       11 100.0%      186 100.0%        3 100.0%       104 52.0%    -34   22   22
  3. Stockfish 11 64 BMI2        400        4 100.0%      300 100.0%       96 100.0%       154 38.5%    -44   17   17

Cross table:
--------------------------------------------------------------------------------------
  #  name                                 score   games         1         2         3
  1. Stockfish 16                     142 71.0%     200         x               142.0
  2. Stockfish 11 64 BMI2 dup         104 52.0%     200                   x     104.0
  3. Stockfish 11 64 BMI2             154 38.5%     400      58.0      96.0         x

Tech:
--------------------------------------------------------------------------------------

Tech (average nodes, depths, time/m per move, others per game), counted for computing moves only, ignored moves with zero nodes:
  #  name                        nodes/m         NPS  depth/m   time/m    moves     time
  1. Stockfish 16                 17892K    10242490     40.6      1.7     69.2    120.8
  2. Stockfish 11 64 BMI2 dup     26977K    15238064     34.4      1.8     52.8     93.5
  3. Stockfish 11 64 BMI2         27141K    15509692     33.7      1.7     60.9    106.6
     all ---                      23908K    13961837     35.8      1.8     61.0    106.9
 

[/quote]

93% draws between the engine and itself that you see with stockfish11 is still not a level that is close to perfect.
The problem is when you see this type of result between 60+60 time control and 5+5 time control or when you see 99.7% draws between the engine and previous version.

We already saw it with stockfish14 based on your posts
See the following threads:

https://prodeo.actieforum.com/t591-stockfish-14-60m60s-vs-stockfish-14-5m5s
https://prodeo.actieforum.com/t445-stockfish-14-vs-stockfish-13-32-threads-tc10m5s-1000-games

Mclane wrote:

Yasser seirawan talks about ideas. But do chess engines have ideas ?!

I thought CSTal was about playing interesting. Not about playing the strongest.

Its IMO almost unimportant if it has a high elo or not, as long as it plays spectacular chess.

Of course doing that against the top 5 is difficult.
Against the top 10 it is possible.
And against the 11-20 it should be possible.

Against human beeings even more.

Yes, the ideas explained in this video have been familiar to chessprograms for ages (this is basically doeable with PSTs and some idea of the concept of development). So a chessprogram could have an idea like: this way I can reach a central majority which is a good thing ceteris paribus.
As far as I understand things (which means: in a very basic way) all evidence suggests that NNUE engines have way more chess knowledge than HCE ones, it’s just that we don’t really know what it is exactly. And so far, the engines can’t explain their thought process to us in a digestible way (which would be a major improvement to their usefulness).
But your idea that most engines just move their pieces around in a random way until they see a mate or a material win is very simplistic and wrong IMHO.
The obvious question is: what is interesting chess? If I look at games Morphy-NN for a while I get a little bored. Yes, the sacrifices can be pretty, but the level of defense is just too miserable. The same is not true for the games of Mikhail Tal – yes, they are way over my head at times, but they clearly have way more to offer concept-wise.
I love to look at games of the times of Alekhine or Lasker. I understand a lot, as this is about my level of understanding chess concepts. I can understand more modern concepts, too – if someone explains them to me as clearly as Yasser does.
Watching games of current Stockfish is pretty useless for me btw. Yes, I can see that what it does works somehow, but there are just too many moves where I have no idea what they are about. I am just too weak a chessplayer to get their beauty. Probably Magnus Carlsen will feel differently.

„ Watching games of current Stockfish is pretty useless for me“

Machine chess kind of Stockfish IS pretty useless indeed.
Therefore i think we will see a renaissance of CSTal, Gambit Tiger or BenjaminII engines that play „different“ then the normal machine chess we see today.
The purpose of chess engines was to satisfy the customers. Not mainly to be used for online games on chess base servers to show: my hardware software combination kills your hardware software combination.

„ But your idea that most engines just move their pieces around in a random way until they see a mate or a material win is very simplistic and wrong IMHO.“

Of course they do not move the pieces arround in a ramdom way.
They had HCE and the programmers taught them how to play good chess.
This had something to do e.g. with the experience the programmers made on championships . They thought i have a strong hadware, so i should win the championship. But then the engine moved the queen on b6, took pawn on b2 and was later trapped and the programmers had to finetune for not moving so much and early with queens, and not get their pieces trapped and better develop before doing these things.

Or they had a won situation on board and the engine was not capable to mate with KNBK.
So the programmers and the programming teams learned out of negative experiences on championships and events.

But then happened something very negative IMO.

The hardware made such a huge improvement that the software was not anymore so important and with a better hardware even stupid engines could make it.

Programmers began to put knowledge out because it made the engine slow.
They decided to use the fast hardware with a slim program to compute searches deeper then the competition.

Uri Blass wrote:

mwyoung wrote:

Chris Whittington wrote:

Admin wrote:

Mclane wrote:

mwyoung wrote:

Mclane wrote:

„ so I suspect it is high because we are close to have some unbeateable engine in chess“


I doubt that such a thing exists,.
Maybe when you have outsearched chess.
But an unbeatable chess engine ?!

CORRECT!

The chess engines draw today, Because the are basically the same with the top engines of chess. All using a computer generated evaluation.

And the chess engines do not know how to beat itself. Or do they understand their weaknesses, or even being able to spot them.

Chess engines are far from perfect. And I have proven this many times. By showing them being unable to solve many chess positions. Even when the solution is well within their search.

Remember Type B chess engines were never designed to play perfect chess, or to be unbeatable. Type B chess engines were design to beat MAN!.

This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out.





I’m with Ed on this. Many engines are similar but that’s not the reason for the draw rate. SF and LC0 are about as dissimilar as can be, but they draw mostly against each other. Komodo has some imbalance knowledge built into it, so it is not exactly the same as SF, but it also has a high draw rate. Computer chess winning is about getting a small advantage, increasing it to a tipping point where it’s then a win. Deeper searches (with a decent eval) usually are able to find moves that avoid any tipping point until the game runs out due to material trades.

Chess is not an unstable game (unlike, say, Shogi). Small advantages do not exponentially run into large ones because material trades down as the game progresses

Quote :

Especially if you have the situation that 1 open source engine is there and strong and all others can begin from this platform of sources.

In older times it was crafty as beginning platform and before that it was gnuchess.

In commercial computerchess age all sources were closed and pretty less was known.

So the engines were completely different.
Mephisto III is completely different to MM2 or Rebel.
Or Sargon or Kittingers engines.

People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other.

Let us disprove this nonsense.

First we will address this fact I posted, and was disputed about TCEC.

From TCEC

Q:  To what degree do you bias openings?

A: In each season’s Superfinal Jeroen has a free hand to do whatever he thinks will result in an interesting and varied contest.  Sometimes his book-exit evaluations will exceed +/-1.00; sometimes he will offer speculative gambits.  His goal is to keep the draw-rate in the 65-80% range each season without a surfeit of one-sided openings.

And even TCEC agrees with our position. Draws are caused by closely match chess engines.

"if very strong, closely matched programs are playing at long time controls with no bias that invites a draw-rate approaching 90%."

I am not sure were the myth started that draws somehow indicated near flawless play by the chess engine. The fact is this has never been true. Draws happen because a chess engine with the same rating. Does not know how to exploit the errors of a similar chess engine in rating and or functions.

Now we will directly disprove this statement.
 
This is my point too.
The engines are not unbeatable . The draw rate happens because they are very similar.

What ?!

Nonsense.

Draws happens because engines make less and less mistakes.

I am out. --- Ed.


Chess engine A plays Chess engine A duplicate. The draw rate scored is 93%

Now was the draw rate caused by the chess engine making less and less mistakes. Or because the chess engine does not know how to exploit the errors of the other chess engine?  

And how can we prove this to see who is correct.

Simple. Have chess engine A play a stronger chess engine. If the draws were caused by mistake free chess. The draws should remain the same.

Lets see what happens....

Chess engine A now plays Chess engine B. Draw rate is now 57%. And chess engine A is crushed.

We will end this with a Mclane quote describing this wrong conclusion reached by circular logic.

"People get the impression that a chess engine like Stockfish is god like because in rating lists mainly modified Stockfish versions drew against each other."

This demonstration was played with Perfect Book 2021 to six moves. And played in gauntlet mode.

Code:

Result:
--------------------------------------------------------------------------------------
  #  name                      games            wins           draws          losses           score    elo    +    -
  1. Stockfish 16                200       85 100.0%      114 100.0%        1 100.0%       142 71.0%     78   26   25
  2. Stockfish 11 64 BMI2 dup    200       11 100.0%      186 100.0%        3 100.0%       104 52.0%    -34   22   22
  3. Stockfish 11 64 BMI2        400        4 100.0%      300 100.0%       96 100.0%       154 38.5%    -44   17   17

Cross table:
--------------------------------------------------------------------------------------
  #  name                                 score   games         1         2         3
  1. Stockfish 16                     142 71.0%     200         x               142.0
  2. Stockfish 11 64 BMI2 dup         104 52.0%     200                   x     104.0
  3. Stockfish 11 64 BMI2             154 38.5%     400      58.0      96.0         x

Tech:
--------------------------------------------------------------------------------------

Tech (average nodes, depths, time/m per move, others per game), counted for computing moves only, ignored moves with zero nodes:
  #  name                        nodes/m         NPS  depth/m   time/m    moves     time
  1. Stockfish 16                 17892K    10242490     40.6      1.7     69.2    120.8
  2. Stockfish 11 64 BMI2 dup     26977K    15238064     34.4      1.8     52.8     93.5
  3. Stockfish 11 64 BMI2         27141K    15509692     33.7      1.7     60.9    106.6
     all ---                      23908K    13961837     35.8      1.8     61.0    106.9
 

93% draws between the engine and itself that you see with stockfish11 is still not a level that is close to perfect.
The problem is when you see this type of result between 60+60 time control and 5+5 time control or when you see 99.7% draws between the engine and previous version.

We already saw it with stockfish14 based on your posts
See the following threads:

https://prodeo.actieforum.com/t591-stockfish-14-60m60s-vs-stockfish-14-5m5s
https://prodeo.actieforum.com/t445-stockfish-14-vs-stockfish-13-32-threads-tc10m5s-1000-games
[/quote]

It does not matter if the draw rate is 93% 95% or 100%. Draws tell you nothing about if a chess engine plays perfect chess.

That is the point.

When most of the top engines are at the same level. You expect a high draw rate.

But guess what

All this nonsense was said about Stockfish 11 and before. The chess engine is unbeatable it does not lose. It must be playing perfect chess.

And this is only true unti we develop stronger chess engines. Then the draw rate crashes.

Only to repeat this circular logic nonsense with the new stronger chess engine.