Author: Vincent Diepeveen
Date: 14:42:13 04/30/01
Go up one level in this thread
On April 30, 2001 at 11:35:56, Aaron Tay wrote: >On April 30, 2001 at 11:21:49, Vincent Diepeveen wrote: > >>On April 30, 2001 at 10:59:03, Ed Schröder wrote: >> >>>On April 30, 2001 at 10:29:14, Vincent Diepeveen wrote: >>> >>>>On April 30, 2001 at 10:01:09, Robert Hyatt wrote: >>>> >>>>>On April 30, 2001 at 07:22:24, Alvaro Polo wrote: >>>>> >>>>>>Hello all, >>>>>> >>>>>>In a recent interview Kramnik states that "We are in a very interesting phase, >>>>>>when the strength of the best GMs and that of the best chess engines run by the >>>>>>best processors are about equal." >>>>>> >>>>>>I know that this point (machines being GM strenght or nor) has been debated >>>>>>again and again and I don't intend to post a troll. I would just like to know if >>>>>>the consensus now among chess programmers is wether Kramnik is right or not. For >>>>>>instance, I remember Bob Hyatt saying that computers are really 2450, etc. But >>>>>>software evolves, CPU power evolves and perhaps now there is agreement that >>>>>>machines are finally GM strenght? >>>>>> >>>>>>Thanks. >>>>>> >>>>>>Alvaro Polo >>>>> >>>>> >>>>>I personally think my estimate is still pretty close. Computers have two >>>>>serious problems: >>>>> >>>>>1. opening books. They depend on a human to "play the game" of choosing good >>>>>and bad openings. This leaves them highly vulnerable to opening preparation and >>>>>traps. Particularly when you practice against one copy and then play another >>>>>copy which doesn't have the 'learning' from the practice games. >>>>> >>>>>2. blocked positions and slow build-ups in kingside attacks. Hardly anyone >>>>>has made progress in fighting either of these problems. And they _still_ offer >>>>>good chances for a GM that is willing to employ them. >>>> >>>>I completely agree here. Add to that that if you can remember which program >>>>you practiced against cq studied very well, that you then also know what >>>>kind of positoinal knowledge the prog doesn't know. >>>> >>>>Some progs lack simplistic positional knowledge. Playing progs in general >>>>is not such a cool idea, you need to know indepth knowledge of the program >>>>you play to win from it. >>>> >>>>At least with 2285 national i need to know that. >>> >>>Do you want to play a match Vincent :-) >>> >>>Ed >> >>What was it, 250$ for each point scored versus rebel? >> >>Under those conditions of course! > >All joking aside, I think It would be very interesting. Rebel has played against >GMs including GMs well known for anti-computer style, but it will be interesting >to see how it stacks up against a weaker player abeit one who really understands >and writes computer Chess programs. How many more rating points do you think >being a computer chess programmer helps? > >The Problem is would playing such a match generate enough publicity to make it >worth for Rebel to do the match? Oh well let's be clear here first about my own level. I'm not such a strong chessplayer. I'm a strong amateur at best. Obviously a first demand would be the level to be 40 in 2 , 20 in 1 + 30 rest of game. Just like many people who work during the week i play at saturdays when national competition is there around 2450 level. I beat IMs normally spoken and lose from GMs. Of course many players play way better at saturday, so in national competition that compensates. When i would play myself a match versus a computer program my level in no way can be seen higher as 2450 therefore. Most likely i do not achieve 6 games in a 6 game match this level. More likely i achieve the rating which i am worth on average FIDE rated. My fide rating can't get up fast as i play too little games for my fide rating. Yet i jump up every list till it's about 2350 i guess. Right now i jumped up from 2150 to 2285. But progress of FIDE rating will get slower probably from now on so to get to 2400 FIDE rating will take a long period of time. My own hope is to get IM before 2005. list of januari 2002 i hope to have my FM. Obviously i hardly beat GMs. Actually i can't remember last GM i have beaten in a slow game. Next year in masterclass i will face again loads of IMs and GMs so that'll be again a good compare. Most likely i'll win again from the IMs and lose from the GMs. My play therefore in the human league is very consistent and definitely somewhere between 2285 and 2450 in any match. The unknown factor however always is the computer. When looking to its tactical strength it definitely has like 3300 nowadays. It's tactical *unbeatable*. Where i won the first few years in tactical ways from the computer i can completely shake that now. In contradiction. If i lose a game it will be because i missed something tactical! Many people however have huge problems to distinguish between tactics and positional play. Especially most GMs do not even know the difference between positional play and strategics. They also can't name the difference between a good bishop and a position bishop. The term position bishop is even unknown in literature, but i mean a KID bishop at d7 for example. "this bishop is worth a rook, but not a position bishop!" most claim loud then. Well... ...give me a rook then... In short it's hard to define in a few lines here the difference between positional, strategics and pure tactics. If i lose most likely because i 'miss something'. Not something tactical, but a positional thing which gets on the board by tactical forced means! Like creating a passed pawn by threatening some things on a certain wing which take care a certain square is not covered very well. Then pawn break and computer has a cool passer. Opponent modelling. Let's go to that weird thing. Opponent modelling is a big joke when you talk about the computer. If a computer is evaluating assymetrical, then it has some weaknesses where i can profit even harder from. What i didn't touch yet is the questions: What is the strategical strength of most computers? What is the positional strength of most computers? Well to start with the last. The easy way to win from a computer like in 1992, those days have gone. King safety is improved in all programs, even the preprocessor type programs. Correct me if i'm wrong, but Fritz even *lost* in DCCC2000 a game against Tao because it sacraficed a pawn too much to get half open files to the king. In short, positional patterns are used to create a strategical won position. In 1992 i won games in blitz usually by giving it loads of material then win. In 2000 and 2001 most games which i win i do NOT give it material soon. An opening where i have won a lot of games with against computers, also in blitz was: 1.e4,e5 2.nf3,nc6 3.c4?,nf6 4.Nc3,bf5 5.h3 and then i started buildup with g3 Bg2 O-O Kh2 d3 and one day f4 f5 g4 g5 and a simple win. Of course this method i never used in 40 in 2 games, for the simple reason that i never risked there to open the game with the dangerous 1.e4 against the computer! closing position with 1.g3 is way easier. Known anti-computer systems like 1.d4 followed by 2.c3 nowadays also challenge even the best computers. Winning such a position is nowadays tougher though. DRawing a peanut. The computer never has a plan if you don't give it an obvious positional advantage. If you give the computer a passer, then it'll win usually. If you give the computer a king side attack, then it mates you. If you give the computer an open center, then it will attack everything of you and win. In short, the ways to get a won position are still numerous, but one must be very realistic acknowledge that even a won position for an amateur like me is impossible to win from the computer if the computer plays the rest of the game at 3300. Another method which i used in the past to get a won position also is slowly dissappearing. What i did was simply 'guess' that my theory preparation was way better as the openingsbook of the program. hoping it would play an outdated line somewhere in a known mainline, i could then get easily a won position. If you are bigtime won and you are capable to not blunder a piece, then you must win of course. No matter whether you're 2200 rated or 2700 rated. Books have improved bigtime last 3 years from commercial programs, also there is more variation in the opening book lines of the programs. So this option to win a game is still valid if i have the luck to get in a line where i know the latest theory from. Like with black i play Slav defense. A line to easy win with with black is the line where white plays a quick Ne5 and then after some sacrafices at e4 you have a bunch of passed pawns versus the computer a piece. In blitz you lose that tactical usually with black, but till today no program knows how to stop 3 connected passers! (note rebel is one of the best to know how to stop passers, so that would be my last choice to play it against, but still i would go for it against Rebel!) Passers are very hard to see tactical consequences from. The only way to do it right is to take into account attacks of opponent pieces. Very little programs have attacks in their evaluation, not to mention good 'block' code. Blocking a single passer or 2 passers is quite different from stopping 3 passers! So in this case i win a game in an endgame without problems. The 3300 tactical level is simply not useful in those positions as the knowledge in the program isn't strong enough to beat it. So then a weaker chain must be seen as the programs strength. In such a position it is more like a 1900 player who doesn't make tactical mistakes. Now of course most players with 1900 make regurarly tactical mistakes in a game, especially endgame as they're exhausted after some hours of play, where i enjoy the moment and only get exhausted when the game is nearly ended, because i start already with a won position when i have an endgame with 3 passers versus a piece. So there are for amateurs big risks in playing mainlines, but also big gains if you have the luck to get in a mainline where you know theory better. Not to mention grandmasters who invent themselve theory and who know way more theory as they are fulltime busy with chess. For me safest option is to play lines where the position gets blocked. when i played crafty some months ago at the internet i lost blitz game after game in blocked positions, as it always managed to open it quick and then it had a won position. This happened till i talked with Bob. He explained to me that he assymmetrically penalizes the position bigtime if there are 8 pawns on the board. Now when i learned this i soon figured out that it was easy to get a won position versus crafty. Because the only thing i had to take care for was to play into lines where opening the position was bad for the computer. A computer throws dices then for a move or 15 and then it opens the position. When it does so i directly have a better position then! A blitzgame later i learned that i could even profit more from this. By both castling the other side, or simply delaying my own castling or keeping king in the center, i could easily launch an attack against crafty's king, as it would happily exchange pawns to open its kings position, giving me cool (half) open file(s) against crafty king. So here you see clearly how important it is to know the style of the program for to profit from it. Some players, even very strong players, completely depend upon their preparement against the opponent nowadays. A good example is a talent at my club: Erwin l'Ami. age : 15 rating : 2327 software: many databases, many chessprograms He wins from all players he knows well, *whatever* their rating. The way in which he does that is by playing openings which he understands and has studied but which are not known by the opponent. All IMs 2400+ he played against which he knew, he has beaten last year. All players he played in masterclass this year he didn't know he lost from! He'll grow in strength so then his strength will take care he wins anyway, also what most people don't realize is that when you get GM you never play a higher rated person who you do not know. Most players you already played a lots of times against. The higher your rating get, the bigger the chance that you already played your opponent before. If you do not know your opponent, then most likely you can also win blindfolded from him. Knowing your opponent is very important. Especially when playing the computer. In general computers have weaknesses, which knowing them is usually enough for real strong players to beat them. For me that's usually not enough. In depth knowledge definitely is a big advantage. One of the biggest preparement mistakes in history of chess is of course made by Kasparov. Several games deep blue showed passive play and happily exchanged queens. In game 6 Kasparov made, as he short after the match said, made a mistake. He played a completely lost line, winning a piece doing that. Probably his first wrong assumption was to guess that the computer would play passive and allow him to develop pieces (after which kasparov would be up a full piece). The second mistake he made was to assume it would probably exchange the queens without problem. This guy (and probably his advisors also must be blamed for this) kasparov definitely will never learn how computer logics work. Every programmer here and most likely a zillion more people will be able to tell that there are rules in nearly *all* programs that if big material down that you must not exchange queens. Also kasparov's king was in the center. Most programs know that (half) open files to the opponent king are very important. Amazing that such a great world champion made such a preparement mistake. Such mistakes even hardly happen to me in blitz where i definitely experiment a lot! Last blitzgame when playing for money (first price us$700) i really felt looking like a fool was a few days ago in the HBI blitz tournament finals against Vaganian who quickly shut in my queen and won within 16 moves or so. Best Regards, Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.