Author: Ross Boyd
Date: 05:06:54 08/26/04
Go up one level in this thread
On August 25, 2004 at 08:55:44, Uri Blass wrote: >On August 25, 2004 at 08:28:54, Ross Boyd wrote: > >>On August 25, 2004 at 07:00:56, Uri Blass wrote: >> >>>On August 25, 2004 at 06:42:47, Vikrant Malvankar wrote: >>> >>>>Your Rating List >>>> >>>>2700 ... World Champion >>>>2550 ... Average GM >>>>2400 ... Average IM >>>>2200 ... National Master >>>>2000 ... Strongest Computers (this was 1985 remember) >>>>2000 ... Candidate Master >>>>1550 ... Average Club Player >>>>1000 ... Average social player >>>> >>>>Well based on the testing done by me >>>> >>>>Commercial programs on P-IV 2.0, AMD ATHLON 2200 (Fritz, Junior,Shredder) >>>>approximately score about 2650-2700 are as good as Average GM and World Champion >>>>in open unbalanced positions. but remember this applies to playing among >>>>themselves only. As far as playing a strong GM well computers have yet to learn >>>>a lot I consider them abt 2550 playing on the aove Harware playing at normal >>>>controls against Super GM. >>> >>>Based on what data? >>> >>>The data that I read suggest better rating and there are >2700 performance of >>>Tiger on slower hardware or of old Junior at durtmond near 2000(it used faster >>>hardware than Athlon2200 but worse software). >>> >>> If u increase the speed things definitely vary but >>>>not the knowledge (chess) in the programs. Thats why Deep Blue lost the match to >>>>Kasparov in 1997. >>> >>>Except facts that are different when deeper blue won in 1997. >>> >>>> >>>>Todays free programs e.g. Crafty, Ruffian are also quiet strong and tactical >>>>monsters. On above hardware at Blitz Time Controls they are abt. 2650. >>> >>> >>>What evidence you have. >>> >>>At blitz time control Fritz3 did better in 1994 on a slow pentium(I am not sure >>>if P90 or even slower hardware) and won a tournament together with kasparov) and >>>Crafty or Ruffian on fast hardware are clearly better than Fritz3. >>> >>>Genius3 did even more than 2650 performance on p90 or p120 in 25 minutes/game in >>>1994 and the free programs on fast hardware are clearly stronger than it. >>> >> >>Another factor may be that Kasparov simply under-estimated Fritz3, now he shows >>computers much more respect. >> >> >>> >>> but >>>>normal Time Control they are 2500 only. >>> >>>Again you need evidence. >>>It seems that you tend to underestimate computers >> >>Maybe we engine authors over-estimate? I don't know. >>I see engines making very deep combinations with ease. I think if we can >>eliminate positional blunders, the type that GM's capitalise on, then computer >>programs can really make some progress. (If only it was so easy). >> >>> >>>The second possibility is that humans improved significantly since 1994 but >>>again you need evidence for it. >>> >>>Uri >> >>Improved their strategy/technique against computers? I'd say they have. Can't >>provide evidence, but it stands to reason, humans are very adaptive. >> >>BTW, has Movei played many rated humans? I'd be interested to hear your results. >> >>Ross > >It only played some correspondence games against weak humans with no impressive >results but I also did not give it a lot of time to give the humans chances(I >gave it only 42 seconds on A1000 for a move when the humans had 2 days per >move(I believe that the humans did not use computers and they were only weak >players with rating of 1600-2200 when most of them were 1600-1900). > >It scored 3.5/6 in a tournament in the israeli chess forum >It lost against >dany salomon 2186 >matan partman 1981 > >won against >yan galvord 2000 >alon shemes 1703 >micheal alkin 1638 > >drew against >liran bas 1746 > >Here is it's last game when it lost against matan partman with rating 1981 >Note that movei could avoid Nd5 with slightly more time on the clock but I >decided that it is unfair to give it more time because humans are so slow when >machines are so fast. > >I also think that 7.Bb5+ is not the best book move but it did not have a big >book and was out of book after some moves. > >[Event "?"] >[Site "?"] >[Date "?"] >[Round "-"] >[White "?"] >[Black "?"] >[Result "*"] > >1. e4 c5 2. Nf3 e6 3. Nc3 d6 4. d4 cxd4 5. Nxd4 Nf6 6. Bg5 Be7 7. Bb5+ Nbd7 >8. O-O a6 9. Bc4 O-O 10. Bxe6 fxe6 11. Nxe6 Qa5 12. Nxf8 Kxf8 13. Qd2 h6 >14. Bh4 Qc5 15. a3 Nb6 16. b4 Qc6 17. f4 Nc4 18. Qe1 b5 19. Nd5 Nxd5 20. >exd5 Bxh4 21. dxc6 Bxe1 22. Rfxe1 Ra7 23. Kf2 Rc7 24. a4 bxa4 25. Rxa4 Nb6 >26. Ra2 Nd5 27. Rb2 Nxf4 28. Rf1 Ke7 29. Re1+ Kd8 30. Re4 Rf7 31. Rb3 d5 >32. Ree3 Kc7 33. Re8 d4 34. h4 Be6 35. Ra3 Bc4 36. Kg3 Ne2+ 37. Kg4 Nc3 38. >Kg3 Kxc6 39. Rc8+ Rc7 40. Rb8 Nd5 41. Rf3 Rd7 42. Rff8 Ne3 43. Rf2 Nf1+ 44. >Kg4 Be6+ 45. Kf3 Rf7+ 46. Ke2 Bc4+ 47. Ke1 Rxf2 48. Kxf2 Ne3 49. Rc8+ Kb5 50. >Rb8+ Ka4 51. Rb7 Nxc2 0-1 >* Hi Uri, I played over Movei's game against Partman. He played well. I agree the Nd5 gave away the dominant centre pawns. I think Ruffian prefers R1-f3 after a short think. It shows how tough it is (for any engine) to beat good humans in strategic positions. Humans can see at a glance what is important... and here we are fiddling with piece-sq tables... we need something much better. The first thing that springs to mind is a pre-search evaluation who's task is to categorise the type of position we are about to search and adjust some evaluation weightings in accordance with the thematics of the position. For instance, it would not be hard to recognise a QGD type position where a minority attack is needed. It gets hard when there are so many categories to hand tune. Thanks for sharing your data, I wonder how the games would have finished if Movei had equal time to think.... Ross
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.