Author: Ed Schröder
Date: 06:28:33 06/25/98
Go up one level in this thread
>Posted by Amir Ban on June 25, 1998 at 05:04:46: >In Reply to: Rebel10's anti-GM revised... posted by Ed Schröder on June 25, >1998 at 03:00:38: >On June 25, 1998 at 03:00:38, Ed Schröder wrote: >>I picked another example to prove my point. Also a tactical one for >>reasons of clearness. Rebel10 anti-GM is NOT about tactics but sometimes >>it certainly has some nice side effects. >>I deliberately do not pick a positional example (although I personally >>believe they are MUCH more important) because you can always argue >>about the moves as we all have our different taste. So a pure tactical >>example and this time there can be no confusion about the key-move. >>r2q1rk1/pbppn1p1/1p2p1Bp/7Q/3PN3/b1P5/P1PB1PPP/1R2R1K1 w - - bm Bxh6; >>This game fragment comes from the WCM Munich 1993, Hiarcs - Genius. >>Rebel10 (without anti-GM) finds Bxh6 on ply-7, score 1.19, time 0:27 >>Rebel10 (with anti-GM) finds Bxh6 on ply-5, score 1.23, time 0:01 >>Note that anti-GM is not about tactical tricks like extra extensions, >>just some specific positional knowledge. >>- Ed - >Ed, >Previously while reading this thread I thought I have nothing useful to give you >as advice. Now I understand a bit better what you want to do, I think I have >some. >Your best chance is in DEFENDING. What you really want to do is to provoke an >attack. For Junior, historically this has always been the big point-winner, >though I've never done anything to invite this (except play badly !). >The reasons are obvious if you think about it. An attack against a computer is a >moment of great trial for the master. Every move must be accurate, and he MUST >make something out of it, and of course the clock is ticking. Mistakes are often >fatal. No shuffling of pieces around for 20 moves to lead you into a positional >loss you do not even start to comprehend (Study 1st game against Fritz. Would >you do better ? Don't kid yourself). Besides, when you are defending, at least >all your NPS is working for some useful purpose. In the positions Anand would >like to arrange for you, your CPU may as well be computing your employees >salaries for all the good it's doing. >Compare your game against Kossashvili in the Aegon to mine (published in >ICCAJ). >Maybe you shouldn't have lost your game, but you did. Maybe I should have lost >my game, but I didn't. Besides, in which type of game do you think Yona felt >more comfortable ? I can ask him if you wish. >To take a more famous example: 6th game of the 1st DB-GK match. There was a >position where eveyone was thinking of Kasparov's possibility to sacrifice in >Bxh7+. Everyone except Kasparov, of course. He didn't want to do it, and he was >right. He would not like to be forced into such a line. >I think it's great that you found this positional term. But maybe you want to >change the sign ? >If you can get Anand to play moves like Rxe6 or d3, you are in good shape. >Amir Well spoken, I loved every word of it! Next it hits several important points of the discussion which aren't touched yet. It's my opinion (mainly based on the Aegon tournaments) that playing against a human till 2200 Rebel (and other programs) mostly easily win. Somewhere and somehow they pick the initiative and win. It's kind of a fixed pattern with very few exceptions. Then in the 2200-2400 area there (mostly) is a fight to get the initiative or a better position. Humans are more clever than the computer in this area but because of minor positional mistakes the computer gets a better position and mostly wins, say 60-65% (?!) Next the 2400-2600 area. Here (in my opinion) I see a total different pattern. Here the human *ALWAYS* somewhere or somehow gets the initiative or the better position. The human simply has a much better understanding about chess and long term plans. Then the long fight begins, the human attacks and the computer defends. Computers (and I totally agree with Amir) are mostly real experts in accurate defending. When the human makes a mistake it is bingo for the computer and Aegon is full of examples of such losses by such strong IM's and lower rated GM's. Last, the > 2600 area leading also to the reason behind "anti-GM". You see he same pattern happening as playing against humans of the 2400-2600 area. The GM builds a strong position, starts an attack and *SELDOM* makes a mistake no matter how good the computer defends. It's an unequal race, the GM is way to good. If the GM has a good attack he will not let the victory slips through his fingers. That's the pattern I saw at Aegon and not only in the Rebel games. Next month Rebel has to play against Anand which is close to 2800 by now. It's my understanding that playing with a "normal" Rebel I have no chance at all keeping the above introduction (to the problem) in mind. It's my opinion once Anand is on the right track and starts an attack Rebel is lost no matter how good Rebel will defend. I have seen Anand playing Rebel at Aegon in the clock-simultan. Rebel lost as a child. Then at the main event of Aegon '97 Rebel played against Kossashvili who later became the tournament winner with an impressive score of 6 of 6. In this game Rebel sacrificed a pawn. Kossashvili already had build himself a good position. Then due to the pawn sacrifice the GM HIMSELF became under great pressure and Kossashvili had a very hard time to escape from all the attacks. Kossashvili finally managed. Rebel lost due to time pressure as Rebel had to go back to the 0:15 level. The loss against Kossashvili would never had happened at longer time controls, at least a draw for Rebel. After the game I talked for a long time with Kossashvili. He said he did not see the pawn sacrifice and he was really impressed by it and what was more important for me he said, "it was the ONLY GOOD move". Thinking about all of this I came to the conclusion that competing on this high level you should make sure your program should not come under such a high pressure because they won't let you go. Maybe on blitz you can escape but certainly not on longer time controls. So you NEED something special that disturbs them. Anti-GM is based on that. Here is the game, the pawn sacrifice is on move 15..Qc2! [Event "AEGON Chess Tournament"] [Site "The Hague NED"] [Date "1997.04.22"] [Round "5"] [White "Yona Kosashvili"] [Black "Rebel 9.0"] [Result "1-0"] [WhiteTitle "GM"] [WhiteElo "2560"] [WhiteCountry "ISR"] 1. c4 c6 2. Nf3 d5 3. e3 Nf6 4. b3 Bf5 5. Bb2 e6 6. Be2 Bd6 7. O-O O-O 8. d3 Nbd7 9. Nh4 Bg6 10. Nxg6 fxg6 11. Nd2 Qc7 12. g3 e5 13. cxd5 cxd5 14. e4 Rac8 15. exd5 Qc2 16. Rb1 Qxd1 17. Rfxd1 Rc2 18. Bf3 Bc5 19. a3 Ne8 20. b4 Bb6 21. Kg2 Nd6 22. a4 a6 23. a5 Ba7 24. Ba1 b5 25. Rb2 Rfc8 26. Nb3 Bxf2 27. Rxc2 Rxc2 28. Rd2 Rxd2 29. Nxd2 Be1 30. Bc3 Nf5 31. Kf1 Bxd2 32. Bxd2 Kf7 33. Bg5 h6 34. Bd8 Ne3+ 35. Ke2 Nc2 36. Kd2 Nxb4 37. Kc3 Na2+ 38. Kb3 Nc1+ 39. Kc2 Na2 40. Kb2 Nb4 41. Kc3 Na2+ 42. Kb3 Nc1+ 43. Kc2 Na2 44. Bc7 Ke7 45. d6+ Ke8 46. Bb6 Nb4+ 47. Kc3 Na2+ 48. Kb3 Nc1+ 49. Kc2 Na2 50. Be3 b4 51. Kb3 Nc3 52. Bc6 Kd8 53. Kxb4 Nb8 54. Bb6+ Kc8 55. d7+ Nxd7 56. Bxd7+ Kxd7 1-0 Rebel10 still will play the good 15..Qc2! but the score of the logical re-capture 15..Nxd5 is still close to the score of 15..Qc2 which easily might cause Rebel to switch to 15..Nxd5 after all. Rebel10 using anti-GM will make the score gap between Qc2! and Nxd5 a lot wider which makes 15..Qc2! guaranteed. - Ed -
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.