https://ratings.fide.com/profile/13500481
Kasparov
Why do people play the benoni defense?
Sort:
Mar 1, 2020
0
#42
Mar 2, 2020
0
#46
dpnorman wrote:
I'm not convinced this is a serious thread but I'll give a serious answer anyway.
I know a decent number of Benoni players. The one who comes to mind first is an FM from my area who plays it every game against d4/c4. People know he does it, but he's seen everything and has good results. It allows him to play asymmetrically against lower-rated players. It gives him a great deal of potential counterplay: the b5 break or otherwise play on the queenside, the possible kingside breaks including f5, active pieces, positions based primarily on understanding, etc.
The idea that the Benoni is just a bad opening is very silly. There might be some white paths to an advantage but realizing that advantage in practice is very difficult. And in many lines, the computer gives +0.9 for a while, and you follow its line, and suddenly you realize it's changing its mind and you're not even better anymore. Or half the time it still thinks you're equal or slightly better, but black has all the piece activity, often via sacrificing something.
It's an annoying opening for many 1. d4 players to face in practice, and I don't even think it's all that bad objectively. I disagree with most of your premises, including the assumption that the number the computer gives, especially after so few moves, means much of anything.
Thank you. Then why is the benoni not fashionable at the top level?
Mar 2, 2020
0
#47
SIowLearner wrote:
It's worth noting that chess evaluations can (and will) change over the course of the game.
Also, Stockfish isn't perfect—it sometimes incorrectly evaluates. Sometimes terribly.
Look at its evaluation scores whenever it loses a game in the TCEC—its evaluation always seems to think it's doing well, or winning ... until it's not. Then the evaluation flips, once it realizes it was wrong.
Stockfish's developers give high eval scores to protected pawn chains that are in enemy territory. In this case, there's a big plus in the eval score because the d5 pawn is protected, and it restricts black's queen knight development.
But black's queen knight doesn't want the c6 square, anyway, so that's a moot point.
Benoni is definitely harder to play from the black end (much like the Sicilian), but an experienced Benoni player welcomes the decreased mobility in exchange for the tactical imbalances, and the huge latent power of black's king bishop.
Interesting, thank you.
I doubt most black players want to play the Benoni Defense as much as want white to be placid with black's gambit which is neatly sideestepped by white forcing black into the Benoni defense.
Utopia247 wrote:
dpnorman wrote:
I'm not convinced this is a serious thread but I'll give a serious answer anyway.
I know a decent number of Benoni players. The one who comes to mind first is an FM from my area who plays it every game against d4/c4. People know he does it, but he's seen everything and has good results. It allows him to play asymmetrically against lower-rated players. It gives him a great deal of potential counterplay: the b5 break or otherwise play on the queenside, the possible kingside breaks including f5, active pieces, positions based primarily on understanding, etc.
The idea that the Benoni is just a bad opening is very silly. There might be some white paths to an advantage but realizing that advantage in practice is very difficult. And in many lines, the computer gives +0.9 for a while, and you follow its line, and suddenly you realize it's changing its mind and you're not even better anymore. Or half the time it still thinks you're equal or slightly better, but black has all the piece activity, often via sacrificing something.
It's an annoying opening for many 1. d4 players to face in practice, and I don't even think it's all that bad objectively. I disagree with most of your premises, including the assumption that the number the computer gives, especially after so few moves, means much of anything.
Thank you. Then why is the benoni not fashionable at the top level?
Because it's by nature rather risky and double-edged. Among other things it means if you're a Super GM, the novelties your Super GM opponents might prep against you could be very dangerous/cutthroat. And the losses will pile up with the wins at a level like that, rather than solidly making half points all the time with black in the Nimzo or Grunfeld. It's also not the sort of opening that suits everyone's style.
For all it's worth, the future of chess engines will be neural network engines. -Also known as artificial intelligent engines, but neural network is a far more accurate description, since they're not actually "intelligent".
But regardless of how it is programmed, they use GPU processors to do the computing, and DO NOT EVALUATE POSITIONS IN TERMS OF CENTIPAWNS. They don't even do alpha-beta paring, since they don't brute force like a normal chess engine does.
That's about all I know on that subject though. They do find plausible lines then play thousands of games quickly in that line, but how they decide is unknown. Some regular chess engines have a monte carlo version that sort of does the same thing, but the mini games it plays internally is still subject to a sort of brute force with alpha-beta paring.
I think the only standalone neural network engine is Fat Fritz at the moment. Leela Chess Zero is distributed.
Kommodo does Monte Carlo style, but I think it's just a modified version of a normal engine. I'm gonna say Monte Carlo looks promising as far as being able to add to chess theory.
It's the Wild West for engines based on the concepts of Alpha Chess Zero, it will be years before it matures, but all of it will increase the horizon of chess theory.
Its a big mistake in that case to trust the engine, if you do a kings gambit or a danish gambit the machine is going to say is wrong as well, the machine considers a perfect play, humans can never do a perfect play, and interesting positions as the ones that we see in gambits or in benoni always works for humans in most levels, maybe the only exception of this is the Elite chess players but that is only if you have over 2700+
santiagomagno15 wrote:
Its a big mistake in that case to trust the engine, if you do a kings gambit or a danish gambit the machine is going to say is wrong as well, the machine considers a perfect play, humans can never do a perfect play, and interesting positions as the ones that we see in gambits or in benoni always works for humans in most levels, maybe the only exception of this is the Elite chess players but that is only if you have over 2700+
I agree with the sentiment of your post. However the engine evaluates the Danish Gambit as a wrong opening in large part because it is a wrong opening.
Why not! How I was introduce to the game
Mar 12, 2020
0
#55
pfren wrote:
Because it is a good opening.
do you have anything else to add?
Mar 12, 2020
0
#57
Caesar49bc wrote:
For all it's worth, the future of chess engines will be neural network engines. -Also known as artificial intelligent engines, but neural network is a far more accurate description, since they're not actually "intelligent".
But regardless of how it is programmed, they use GPU processors to do the computing, and DO NOT EVALUATE POSITIONS IN TERMS OF CENTIPAWNS. They don't even do alpha-beta paring, since they don't brute force like a normal chess engine does.
That's about all I know on that subject though. They do find plausible lines then play thousands of games quickly in that line, but how they decide is unknown. Some regular chess engines have a monte carlo version that sort of does the same thing, but the mini games it plays internally is still subject to a sort of brute force with alpha-beta paring.
I think the only standalone neural network engine is Fat Fritz at the moment. Leela Chess Zero is distributed.
Kommodo does Monte Carlo style, but I think it's just a modified version of a normal engine. I'm gonna say Monte Carlo looks promising as far as being able to add to chess theory.
It's the Wild West for engines based on the concepts of Alpha Chess Zero, it will be years before it matures, but all of it will increase the horizon of chess theory.
Yes, after all it is all about programming techniques. Artificial General Intelligence is not well understood yet, we Don't really know how to give computers what we Could call human traits such as reasoning or perception. Anyway, for chess, one method which was used to program Deep blue is called Heuristics, which is used for very specialized programming. The idea is that it draws the program towards certain parts of the search, and away from others, thus Trying to always make the best choice according to the "Rules" it is given. But, it can make mistakes too, If it is programmed to think that the Queen is more important than a Bishop, then it will react accordingly, and sometimes make a wrong choice. I Don't know about Today's computers, since know computers are able to play against themselves, they probably make decisions based on "experience" more than coding.
Mar 13, 2020
0
#58
Utopia247 wrote:
So WHY are computers misevaluating the benoni, and what would the correct centipawn evaluation be?
The problem is that anybody that is good at chess will tell you that "centi-pawn" evaluation is total garbage. Who cares if it is +0.27 or +0.31? You going to say the former is good and the latter is bad? With that type of difference, you cannot even say the first move is better than the second.
Not all advantages in chess can be directly translated to a point system. Even the Pawn is 1, Bishop is 3, Queen is 9, etc system is artificial, and was strictly built for the total beginner to understand the difference in strength of each individual piece on an empty board. Problem is, chess is not linear.
Case in point. Let's say your favorite food is mashed potatoes and that gets 4 points. Your next is chocolate chip cookies and that gets 3 points. Your next is a cherry slush and that gets 2 points. Your next is chocolate ice cream and that gets 1 point.
Together, the cookies and slush give you 5 points. Not a bad snack. The mashed potatoes and chocolate ice cream are also 5 points, so let's give you that. I will put a scoop on your potatoes! They don't go together, do they?
Chess pieces are the same way. Individual evaluation you can throw out the window. A Queen and Knight often mingle well. They might be better than your 2 Rooks and 2 extra pawns! Both are 12! Does not mean jack!
You have to learn to use the simple assessments like equal, white is winning, black is slightly better (maybe because his game is easier to play), etc.
Forget about computer centipawn values.
Where computers beat humans by a landslide is finding things like a 14 move forcing sequence that wins a whole rook due to various mate threats along the way that result from double piece sacrifices. All it takes is one of those and you are done! That is why computers beat people!
https://ratings.fide.com/card.phtml?event=20697350
BANANA