Author: Robert Hyatt
Date: 14:02:48 01/06/00
Go up one level in this thread
On January 06, 2000 at 10:14:45, Graham Laight wrote: >On January 05, 2000 at 22:38:57, Robert Hyatt wrote: > >>I find this entire discussion extremely funny. >> >>The _programmers_ say the programs are _not_ 2600 players. The non-programmers >>claim they are. Who would you suppose has the best perspective to make that >>judgement? > >Unless they're 2600 players themselves, how would they know? It isn't hard, believe me.. > >Perhaps the non-programmers are weighing the evidence more impartially than the >programmers. The programmers might be too engrossed in the trees to be able to >see the forest! The other way around, in fact. The 'non-programmers' are too caught up watching GMs lose blitz games on the servers, and they pay less attention to the games than they do to the results of the games. many 'won' games are lost by a GM when he makes a small tactical error. Part of the game, you say? Of course. But the computer played into a lost position and lucked out. I see this in _many_ games. I have seen many games where Crafty has simply outplayed a GM move for move, even into endgames (I used to fear reaching endgames with GM players, but in general don't any longer as it has improved an awful lot there). But for every game where it plays like Karpov, there are several games where it plays much worse but manages to win on tactical trickery. At 40/2, this tactical trickery becomes much less likely. Then things change. The 'non-programmers' are ignoring the fact that we have very little data to go on at 40/2hr time controls. So everyone is extrapolating from game/30. But it doesn't work as well as they would hope, IMHO. > >>:) >> >>IE find _any_ programmer here that would say his program is a 2600 level player >>if it was playing in FIDE rated events. Find one. even one-half of one. :) > >If I claimed to he half a chess programmer, would my opinion count? Yes. _after_ you write a chess program, and spend time talking with GM players, you will begin to see just how much the programs of today _don't_ know. Because until you talk with them regularly, you can't appreciate just what they know and are capable of at the chess board. The title "GM" doesn't come easy. > >And another thing - it's no good saying that the SSDF ratings inflate too much >because it's so easy to make adjustments to your program to make it win against >itself. I agree - it's probably easy to do that - but to make it win more often >against the diverse range of programs that exist in the SSDF is another matter. > >-g programs are all very similar. Strong in tactics. Weak in positional understanding. When someone adds something new (outside passed pawns for example) they will win a lot of extra games and their rating will go up. New versions of other programs come out with this feature, and the new group settles in 50 points above the old group without that piece of knowledge. Yet nothing says that new group would do any better against GM players than the original group... That is how rating inflation happens. And you can clearly see the stratification that goes on within the SSDF rating list.
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.