Bill Rogers wrote:I don't know if the man who goes by the name of "Father" is a GM or not but he has demonstrated that any program can be beaten or let us say a way to keep the program from beating him. Can any of you imagine what might happen if a GM used the same logic.
Bil
The so called "Father" uses a stupid bug in the ChessBase GUI to achieve his even more stupid wins on time
His wins has nothing to do with our discussion here,it's simply a monkey trick that he exploites again and again....
How does he do it, how can you take advantage of the bug?
I think he uses the stonewall and time to beat the computer chess.
There are other ways to beat them to.
This is not fair because Pablo is taking advanage of the computer
engines weakness.
bob wrote:Then we are on opposite poles with light-years between us. We also saw Kasparov get good positions against Junior and Fritz and then blow the games. But he did get "good positions". The programs just do not have an evaluation sophisticated enough to compare with a GM's abilities. Unfortunately for the GM, the programs _do_ have tactical skills (partially based on deep search, partially based on consistency) that are more than enough to compensate for the most part. But if you really think that programs understand the minor nuances of pawn structure and such, boy do we disagree...
Bob, the last match Kasparov played was five years ago! Believe it or not the programs of today are much stronger positionally than X3D Fritz was.
I'll take "not"... Tactically they are much better. Positionally the only change is that the deeper searches help some. But not in terms of new evaluation tricks... not much has changed there.
Just because modern programs don't have some breakthrough new evaluation trick does not mean they are not positionally far stronger than X3D Fritz. Just look at the games of a modern program like Rybka versus an old program like X3D Fritz. Frequently Rybka wins by slowly grinding down its opponent.
Uri wrote:Strategy is one of the areas computers are weak at. In 1996, Kasparov crushed Deep Blue in round 6 by demonstrating his superior understanding of pawn play and space advantage. The same was truth about his win against X3D Fritz in this game.
I believe that Kramnik and Kasparov when playing their best chess are stronger than Rybka.
I think that it is the opposite.
Rybka can lose to kasparov or kramnik when she play her worst chess but usually it does not happen.
There are positions that programs do not know what to do but these positions are minority of the positions and usually humans cannot get these positions from the opening positions.
In most positions programs play positionally better than humans and
even if they lose a game against the best players(I do not claim that it is impossible) the game does not contradict my claim.
Uri
"in most positions..." is about the _worst_ statement I have ever seen you write. "In some ..." might be reasonable. But not even "In many..." They will occasionally stumble into a great positional move, but for the most part they do not, and they create weaknesses that would lead to a loss were it not for the human's great tendency to make mistakes.
I think that it is the opposite.
They will occasionally make positional blunders but in most cases they play better than humans positionally.
Your theory was right some years ago and I can say that kasparov got positional advantage against deeper blue that he could not translate to winning the game in some games of the match(for example games 4 and game 5) but today it is not the case that humans get positional advantage against machines and if you watch hydra-adams games you can see that hydra got positional advantage and there was no situation when adams got the advantage that means a situation that hydra had to defend inferior position to get a draw or to win thanks to a tactical mistake.
Uri
Then we are on opposite poles with light-years between us. We also saw Kasparov get good positions against Junior and Fritz and then blow the games. But he did get "good positions". The programs just do not have an evaluation sophisticated enough to compare with a GM's abilities. Unfortunately for the GM, the programs _do_ have tactical skills (partially based on deep search, partially based on consistency) that are more than enough to compensate for the most part. But if you really think that programs understand the minor nuances of pawn structure and such, boy do we disagree...
There was a big progress in computer chess since the matches of kasparov with Fritz and Junior.
Kasparov played with programs that are more than 200 elo weaker than rybka2.3.2a
Even if you assume that the only relative advantage of rybka is better search
I doubt if kasparov could do the same against a version of Junior or Fritz that is 10 times faster.
Search help to find better positional moves.
Uri
There has not been any "big progress" in terms of evaluation. Hardware speed, yes. Perhaps search tricks like LMR and whatever different thing Rybka is doing. But not in the "smarts"...
1)We do not know if the reason for progress is almost only better search tricks.
It is clearly possible that there was also a big progress in terms of evaluation.
2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri wrote:Strategy is one of the areas computers are weak at. In 1996, Kasparov crushed Deep Blue in round 6 by demonstrating his superior understanding of pawn play and space advantage. The same was truth about his win against X3D Fritz in this game.
I believe that Kramnik and Kasparov when playing their best chess are stronger than Rybka.
I think that it is the opposite.
Rybka can lose to kasparov or kramnik when she play her worst chess but usually it does not happen.
There are positions that programs do not know what to do but these positions are minority of the positions and usually humans cannot get these positions from the opening positions.
In most positions programs play positionally better than humans and
even if they lose a game against the best players(I do not claim that it is impossible) the game does not contradict my claim.
Uri
"in most positions..." is about the _worst_ statement I have ever seen you write. "In some ..." might be reasonable. But not even "In many..." They will occasionally stumble into a great positional move, but for the most part they do not, and they create weaknesses that would lead to a loss were it not for the human's great tendency to make mistakes.
I think that it is the opposite.
They will occasionally make positional blunders but in most cases they play better than humans positionally.
Your theory was right some years ago and I can say that kasparov got positional advantage against deeper blue that he could not translate to winning the game in some games of the match(for example games 4 and game 5) but today it is not the case that humans get positional advantage against machines and if you watch hydra-adams games you can see that hydra got positional advantage and there was no situation when adams got the advantage that means a situation that hydra had to defend inferior position to get a draw or to win thanks to a tactical mistake.
Uri
Then we are on opposite poles with light-years between us. We also saw Kasparov get good positions against Junior and Fritz and then blow the games. But he did get "good positions". The programs just do not have an evaluation sophisticated enough to compare with a GM's abilities. Unfortunately for the GM, the programs _do_ have tactical skills (partially based on deep search, partially based on consistency) that are more than enough to compensate for the most part. But if you really think that programs understand the minor nuances of pawn structure and such, boy do we disagree...
There was a big progress in computer chess since the matches of kasparov with Fritz and Junior.
Kasparov played with programs that are more than 200 elo weaker than rybka2.3.2a
Even if you assume that the only relative advantage of rybka is better search
I doubt if kasparov could do the same against a version of Junior or Fritz that is 10 times faster.
Search help to find better positional moves.
Uri
There has not been any "big progress" in terms of evaluation. Hardware speed, yes. Perhaps search tricks like LMR and whatever different thing Rybka is doing. But not in the "smarts"...
1)We do not know if the reason for progress is almost only better search tricks.
It is clearly possible that there was also a big progress in terms of evaluation.
2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri wrote:Strategy is one of the areas computers are weak at. In 1996, Kasparov crushed Deep Blue in round 6 by demonstrating his superior understanding of pawn play and space advantage. The same was truth about his win against X3D Fritz in this game.
I believe that Kramnik and Kasparov when playing their best chess are stronger than Rybka.
I think that it is the opposite.
Rybka can lose to kasparov or kramnik when she play her worst chess but usually it does not happen.
There are positions that programs do not know what to do but these positions are minority of the positions and usually humans cannot get these positions from the opening positions.
In most positions programs play positionally better than humans and
even if they lose a game against the best players(I do not claim that it is impossible) the game does not contradict my claim.
Uri
"in most positions..." is about the _worst_ statement I have ever seen you write. "In some ..." might be reasonable. But not even "In many..." They will occasionally stumble into a great positional move, but for the most part they do not, and they create weaknesses that would lead to a loss were it not for the human's great tendency to make mistakes.
I think that it is the opposite.
They will occasionally make positional blunders but in most cases they play better than humans positionally.
Your theory was right some years ago and I can say that kasparov got positional advantage against deeper blue that he could not translate to winning the game in some games of the match(for example games 4 and game 5) but today it is not the case that humans get positional advantage against machines and if you watch hydra-adams games you can see that hydra got positional advantage and there was no situation when adams got the advantage that means a situation that hydra had to defend inferior position to get a draw or to win thanks to a tactical mistake.
Uri
Then we are on opposite poles with light-years between us. We also saw Kasparov get good positions against Junior and Fritz and then blow the games. But he did get "good positions". The programs just do not have an evaluation sophisticated enough to compare with a GM's abilities. Unfortunately for the GM, the programs _do_ have tactical skills (partially based on deep search, partially based on consistency) that are more than enough to compensate for the most part. But if you really think that programs understand the minor nuances of pawn structure and such, boy do we disagree...
There was a big progress in computer chess since the matches of kasparov with Fritz and Junior.
Kasparov played with programs that are more than 200 elo weaker than rybka2.3.2a
Even if you assume that the only relative advantage of rybka is better search
I doubt if kasparov could do the same against a version of Junior or Fritz that is 10 times faster.
Search help to find better positional moves.
Uri
There has not been any "big progress" in terms of evaluation. Hardware speed, yes. Perhaps search tricks like LMR and whatever different thing Rybka is doing. But not in the "smarts"...
1)We do not know if the reason for progress is almost only better search tricks.
It is clearly possible that there was also a big progress in terms of evaluation.
We have a good idea of Rybka 1's evaluation, thanks to Strelka. Nothing unusual there, and in fact, there are signifncant things _missing_. So that shoots down the "significant progress over past 5 years" in terms of evaluation, without doing any additional research.
2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
I don't see anything that suggests that at all. The definition of "singular extensions" is generic. It is how you define "one move is significantly better by some margin S" that controls what gets extended, however. Doesn't have to be tactics, can easily be positional as well.
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri
Again, it is mainly about the search. I'm convinced of that in watching the games. Rybka does have some unusual positional scores, to be sure. I have some I have been looking at from the last ACCA event.
Uri Blass wrote:2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
I don't see anything that suggests that at all. The definition of "singular extensions" is generic. It is how you define "one move is significantly better by some margin S" that controls what gets extended, however. Doesn't have to be tactics, can easily be positional as well.
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri
Again, it is mainly about the search. I'm convinced of that in watching the games. Rybka does have some unusual positional scores, to be sure. I have some I have been looking at from the last ACCA event.
I don't see how you can be sure that Rybka's improved play is "mainly about the search" when you also acknowledge that it also has "some unusual positional scores". And that Rybka frequently wins simply by getting gradually better positions and not by tactical oversights of its opponents seems to me to be fairly indicative of improved positional play. How else can you explain it? Also Rybka often does worse than other programs on tactical test positions yet wins matches consistently. Having a rating more than 200 points higher than X3D Fritz, and yet not doing dramatically better on tactical test suites does not seem consistent with a view that Rybka's positional play is not much better than X3D Fritz's.
Uri Blass wrote:2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
I don't see anything that suggests that at all. The definition of "singular extensions" is generic. It is how you define "one move is significantly better by some margin S" that controls what gets extended, however. Doesn't have to be tactics, can easily be positional as well.
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri
Again, it is mainly about the search. I'm convinced of that in watching the games. Rybka does have some unusual positional scores, to be sure. I have some I have been looking at from the last ACCA event.
I don't see how you can be sure that Rybka's improved play is "mainly about the search" when you also acknowledge that it also has "some unusual positional scores". And that Rybka frequently wins simply by getting gradually better positions and not by tactical oversights of its opponents seems to me to be fairly indicative of improved positional play. How else can you explain it? Also Rybka often does worse than other programs on tactical test positions yet wins matches consistently. Having a rating more than 200 points higher than X3D Fritz, and yet not doing dramatically better on tactical test suites does not seem consistent with a view that Rybka's positional play is not much better than X3D Fritz's.
OK, poor wording on my part. "Positional scores" are scores that don't appear to show any material gain. But Rybka tries to confound analysis by reporting false node counts, truncated PVs and fake search depths. What I was seeing was in a particular case where Crafty was searching very deeply itself, Rybka reported a score of almost winning a pawn 3-4-5 moves before Crafty saw any problem. Rybka's PV didn't show a pawn falling, so I lump that into the "positional scoring" column. Which may well not be correct, because of all the phony information Rybka provides. I'm convinced it saw some deep tactic well before I did, which is a search issue, not evaluation, even though looking at the PV one would assume otherwise.
Your argument sounds good until you go back 12-14 years and look at chess genius, which was beating everybody yet it was not particularly stronger or weaker than others on tactical positions. But his search was seeing things we were not seeing. We now know that part of it was null-move, mixed in with odd/even ply agressive pruning. But it was well above the rest for a few years, yet its evaluation didn't even include some key endgame knowledge such as majorities and the like that others of the day (including Cray Blitz and later Crafty to name mine) did have. Significant search tricks are _far_more likely to produce dominating performance than simple evaluation additions...
Uri Blass wrote:2)Better search is enough to find better positional moves.
It is clear that the better search of rybka is not about forced lines and you can find programs that solve problems faster than rybka(based on analysis with rybka it seems clear that rybka does not use singular extensions).
I don't see anything that suggests that at all. The definition of "singular extensions" is generic. It is how you define "one move is significantly better by some margin S" that controls what gets extended, however. Doesn't have to be tactics, can easily be positional as well.
The main advantage seems to be searching deeper in quiet lines
Things may be connected because if you make big reductions for sacrifices then you can search deeper in quiet lines.
Uri
Again, it is mainly about the search. I'm convinced of that in watching the games. Rybka does have some unusual positional scores, to be sure. I have some I have been looking at from the last ACCA event.
I don't see how you can be sure that Rybka's improved play is "mainly about the search" when you also acknowledge that it also has "some unusual positional scores". And that Rybka frequently wins simply by getting gradually better positions and not by tactical oversights of its opponents seems to me to be fairly indicative of improved positional play. How else can you explain it? Also Rybka often does worse than other programs on tactical test positions yet wins matches consistently. Having a rating more than 200 points higher than X3D Fritz, and yet not doing dramatically better on tactical test suites does not seem consistent with a view that Rybka's positional play is not much better than X3D Fritz's.
OK, poor wording on my part. "Positional scores" are scores that don't appear to show any material gain. But Rybka tries to confound analysis by reporting false node counts, truncated PVs and fake search depths. What I was seeing was in a particular case where Crafty was searching very deeply itself, Rybka reported a score of almost winning a pawn 3-4-5 moves before Crafty saw any problem. Rybka's PV didn't show a pawn falling, so I lump that into the "positional scoring" column. Which may well not be correct, because of all the phony information Rybka provides. I'm convinced it saw some deep tactic well before I did, which is a search issue, not evaluation, even though looking at the PV one would assume otherwise.
Your argument sounds good until you go back 12-14 years and look at chess genius, which was beating everybody yet it was not particularly stronger or weaker than others on tactical positions. But his search was seeing things we were not seeing. We now know that part of it was null-move, mixed in with odd/even ply agressive pruning. But it was well above the rest for a few years, yet its evaluation didn't even include some key endgame knowledge such as majorities and the like that others of the day (including Cray Blitz and later Crafty to name mine) did have. Significant search tricks are _far_more likely to produce dominating performance than simple evaluation additions...
I am not sure if search is the main advantage of strelka or rybka.
The point is that based on our knowledge of the search of strelka significant things are missing.
Strelka have no partial extensions and this is a significant thing that is missing.
Strelka extends too much like Fruit and extend all checks and single reply to check by a full ply(this was improved in later versions of rybka that do not solve instantly problems with long sequence of check and single reply to check)
This mean that it is not clear that Strelka made significant progress in terms of search relative to other programs and maybe the progress is in the evaluation.
