Author: Robert Hyatt
Date: 20:33:30 01/07/00
Go up one level in this thread
On January 07, 2000 at 19:24:45, Bertil Eklund wrote: >On January 07, 2000 at 13:31:39, Robert Hyatt wrote: > >>On January 07, 2000 at 13:23:07, Chris Carson wrote: >> >>>Albert, >>> >>>All ELO ratings are calculated that way (SSDF, FIDE, PCA, ...). >>> >>>They are all based on opponents ratings at the time. :) >>> >>>I missed your point. However, I respect your opinion and agree to >>>disagree. :) >>> >>>So tell me, >>> >>>What do you think the rating of Hiarcs 7.32 is in FIDE? (Not Rebel Century) >>>Why do you think that? Do you consider that a valid and reliable estimate? >>>Why? How confident are you with that estimate? Why? >>> >>>Perhaps I have missed something. Convince me to change my mind, but please >>>do not tell me I am wrong just because you disagree, give me a better estimate >>>and a better way to prove the estimate is correct. I am not looking for >>>the SSDF is bad argument, I am looking for a better estimate that stands >>>on its own and does not need to bash SSDF or someone elses opinion. :) >>> >>>Best Regards, >>>Chris Carson >> >> >>My estimate still has not changed. I think that the top programs would probably >>settle in at 2450 or so, maybe as high as 2500 if they are lucky. But no >>higher. And I think that the top programs would do fairly similarly against >>humans (although I think some of the 'fast searchers' would have serious >>problems once the GM/IM players figured them out). IE on the SSDF list the >>top 10 programs might be separated by a span of 150 rating points. Against >>humans I think this would compress to 50 or so for the most part... > >Hi! > >What on earth is this for nonsense, it doesn´t matter what program you use or >what computer speed you have all programs performs the same level of 2450 or so >against your GM-friends. Have you programmed your friends? > >Bertil >> >>And I think it would steadily drop if the engines remain static. Humans learn. >>Computers don't. Based on my own experiments, faster hardware is much more significant in comp vs comp games than it is in comp vs human games. Based on _thousands_ of games played on the servers. For several months I tried two machines on the same account, alternating... a P5/133 and a P6/200, which was about 2.5X faster. The rating did not change up or down nearly as much as the 2.5x speed-up would suggest (ie nowhere near 100 rating points). In fact, it wasn't uncommon to see both machines with very similar ratings, when only playing against humans... That is my data... I thought it interesting, not surprising, and only ran the test for fun. Crafty's rating, when only playing humans, on ICC reached over 3250 and stayed there for a month. Once the computer guys woke up (No idea what they were doing) and started playing me again, the rating dropped back to about 3000 where it has stuck. That is Crafty playing 250 points stronger vs humans than against computers. Real data, over thousands of games...
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.