Author: José Carlos
Date: 10:14:42 04/08/03
Go up one level in this thread
On April 08, 2003 at 11:48:56, Robert Hyatt wrote: >On April 08, 2003 at 09:48:21, Chris Carson wrote: > >>On April 08, 2003 at 09:12:05, Jouni Uski wrote: >> >>>I included Carson's ratings to latest Fide with interesting result :-) >>> >>>1 DEEP BLUE 97.................... 2862 >>>2 DEEP JUNIOR X................... 2847 >>>3 Kasparov, Gary.................. 2830 >>>4 DEEP FRITZ 8.................... 2807 >>>5 Kramnik, Vladimir............... 2789 >>>6 Anand, Viswanathan.............. 2764 >>>7 Leko, Peter..................... 2746 >>>8 Shirov, Alexei.................. 2735 >>>9 Topalov, Veselin................ 2735 >>>10 Bareev, Evgeny.................. 2734 >>> >>>Please no comments about different rating pools - all ratings are FIDE and >>>from same pool! >>> >>>Jouni >> >>Nice observation and may generate some discussion. Enjoy the posts. :) >> >>Some observations: >> >>Arguments against the comp ratings. >>1. The comps do not have enough games to establish a rating. >>2. The comps listed only played against one opponent under match conditions. >> >>Arguments that favor the comps. >>3. In tournament conditions (human v human and human v comp) the comps score >> about 100 points higher since the human does not prepare just for the comp. >>4. The comps in this list only played the top 2 GM's, against lower rated GM's >> (match or tournament) the comp ratings would increase. > >That suggests that the Elo formula doesn't work. Can you explain why you >feel that way??? > >IE if I can get a _higher_ rating by playing lower-rated opponents, than I can >get by playing >higher-rated opponents, the Elo system is basically flawed. I don't believe it >is, myself. I don't think it's basically flawed, but it's not perfect either. In my experience as a player, I used to get higher rating playing strong players than weak players. I believe the reason was that I had an aggresive but terribly irregular playing style. I was able to beat strong players as easily as was beaten by weak ones. On the other hand, computers don't blunder except for some bug or big lack in eval, which happens almost never to professionals. As a result, a program will consistently beat weaker players without getting tired or making mistakes. This is far from being a proof, it's just a reasoning, and I still don't know that anyone has tried an experiment to prove/refute it. José C. >> >>In my opinion, the above ratings for the comps reflect the performance to be >>expected for match play when the Top Human GM has months to prepare and plan for >>the comps (DF and DJ anyway, not DB). The true ratings for the comps would be >>at least 100 to 200 points higher if they could play in FIDE tournaments. >> >>There will be a few FIDE FM's, IM's and GM's that will play well against the >>comps, however, these games will be few and most FM's, IM's and GM's will get >>clobbered by the comps in a tournament environment when the comps are on 2Ghz >>and faster machines (fastest duals and singles procs with Rebel, Fritz, Junior, >>Shredder, Hiarcs, Chess Tiger, Gandalf and ChessMaster) would take the top 8 >>places in most GM tournaments. Perhaps Ruffian and others could be added to the >>list. >> >>Just my take. :)
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.