Author: Robert Hyatt
Date: 09:24:51 07/25/98
Go up one level in this thread
On July 25, 1998 at 11:01:53, Don Dailey wrote: >On July 25, 1998 at 06:47:27, Leon Stancliff wrote: > >>On July 25, 1998 at 02:00:59, Georg Langrath wrote: >> >>>I think that the time has come when commercial computers must show if they have >>>a chance in slow matches. The day a computer wins 8 matches with time control >>>40 moves i 2 hours against a GM, then we have a new era. >>> >>>Georg >> >>I think it is generally accepted that the top computers are still below the >>level of Kasparov, Karpov and Anand in 40/2 games. The real question is whether >>or not these top programs can play in the 2500-2600 Elo range. If so, they can >>be considered Grandmaster level. >> Your suggestion is valid. However, we need to find one, two or three >>Grandmasters who have about a 2600 Elo rating and are willing to play Rebel 10 >>on the same hardware it used against Anand. Is there a third party sponsor out >>there who would financially support three eight game matches at 40/2? >> I have little doubt that Rebel 10 would be willing. I have considerable more >>doubt that there will be three GM in the 2600 range who will show an interest. >> >>Come on!! Prove me wrong!! > >This is an excellent idea. But first we should play a match with >some players near Rebels level before moving up. Here is my >suggestion: > >Bob Hyatt puts an upper bound of 2400 on the strength of the very >best micro's. If we play a match with some humans, then we target >players close to Rebel's strength to determine if there is any >need to "bother" the good players. > >The idea, is to get players as CLOSE as possible to the >range of strength we think Rebel is. You don't want Rebel >playing up in every game, or visa versa. It requires a larger >sample of games to accurately rate such matches since a single >draw can change your performance rating enormously. > >So I suggest a range of 2250-2450 for our opponents. This is >100 points on either side of 2350. I pick 2350 as center >because if we take 2400 as upper bound, then 2350 might be >considered a reasonble "best guess" estimate. Bob, if >you are listening, does 2350 seem like a reasonable estimate >of Rebel's chess strength? I'm always listening. :) It depends on the time control. I can speak better about Crafty, having watched so many games. Here's my analysis there: blitz (5 mins/game or so) 2800+ somewhat slower (game/15) 2600+ game/30-game/60 ~2500, or maybe slightly lower 40/2 ~2400. This is based on several hundred games I've summarized from ICC results and has to be taken with a grain of salt. I believe the 2800 blitz result because Cray blitz lost 3 blitz games, out of over 120 played, during its last 5 years of competition at various events. The game/15, game/30 results have surprised me.. but they have come against players like Roman, Larry Christiansen, Walter Browne, and so forth, I don't have any 40/2 games against GM's although I have a few against IM players... I tend to think that most of the programs are in the same ballpark when it comes to playing against humans, in general. Some might do worse, others might do better, but on average, I like the 2400 number. I know a couple of IM's that are going to take me to task for this, however... :) > >Now if Rebel does poorly in such a match, it will be an >indication that 2350 is too optimistic a rating for Rebel, >but more importantly, it will tell us we need to wait before >challenging a group of grandmasters, unless we just want to >watch Rebel get embarassed. On the other hand, if Rebel >does well, then we have a serious basis for making a more >aggressive challenge. > >The beauty of this system, is that it will be MUCH easier to >get a bunch of 2200+ players together for this experiment. >And it will tell us whether to continue with stronger players >or not. > >I personally take 2400 to be the lowest believable rating >that a program like Rebel could have, I believe it's likely >to be more like 2500. 2400 is MY lower bound. But since >Bob is more experienced in these matters I would be willing >to defer to him and try this experiment first, especially >since it would be MUCH easier to get it going. I still think >it's too early to go after the GM's. you are probably right. It is hard enough to get IM players to go for this sort of stuff, much less GM players. if we pick the right IM players, those that have "computer experience" it will be interesting... *if* we can get even them to participate. It will not be easy... > >- Don
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.