Author: Robert Hyatt
Date: 18:36:58 04/29/02
Go up one level in this thread
On April 29, 2002 at 18:12:44, martin fierz wrote: >On April 29, 2002 at 18:05:25, Robert Hyatt wrote: > >>On April 29, 2002 at 17:13:31, martin fierz wrote: >> >>>On April 29, 2002 at 16:55:45, Robert Hyatt wrote: >>> >>>>On April 29, 2002 at 16:15:27, Roy Eassa wrote: >>>> >>>>>On April 29, 2002 at 15:50:21, Robert Hyatt wrote: >>>>> >>>>>>On April 29, 2002 at 13:56:58, Roy Eassa wrote: >>>>>> >>>>>>> >>>>>>>How do longer time controls affect humans and computers? >>>>>>> >>>>>>>For humans, the extra time mainly provides better "debugging" of one's analysis. >>>>>>> It also gives more chances to find different lines and greater depth, but these >>>>>>>are quite secondary for human GMs, IMHO. >>>>>>> >>>>>>>For computers, better debugging is (almost) not an issue. They make no tactical >>>>>>>errors within their horizons. What the extra time gives computers is mainly >>>>>>>greater search depth. But doubling the time does not even add 1 ply usually. >>>>>>> >>>>>>>So, which factor makes the bigger difference, GMs getting debugging that's twice >>>>>>>as good or computers getting less than 1 ply of greater depth? >>>>>>> >>>>>>>When GMs lose to computers, it's *almost always* due to insufficient debugging. >>>>>>>Doubling the time (for example) can make a HUGE difference here. >>>>>>> >>>>>>>When computers lose to GMs, it's *occasionally* due to insufficient depth that >>>>>>>could be cured by doubling the time. >>>>>>> >>>>>>>Obviously, both humans and GMs play stronger on an *absolute* scale when given >>>>>>>more time. But I think it's most likely that GMs benefit *proportionally* much >>>>>>>MORE than computers do from the additional time. >>>>>> >>>>>>] >>>>>>It is trivial to test. play some game/1 game/5 game/15 and game/60 games >>>>>>vs the same GM. See what happens. I already know. :) >>>>> >>>>> >>>>>Trivial? Maybe YOU have a human GM lying around your house, waiting to do this, >>>>>but I don't! ;-) >>>> >>>> >>>>Play such a series of games against _any_ human... the resulting curve will >>>>be roughly the same... >>> >>>dear bob, >>> >>>if you have such numbers, could you please post them? there are people here who >>>believe in things like "humans get tired if they think for a long time" and >>>other crazy stuff like that - i have no numbers to disprove their statements, >>>but i know they are wrong. do me a favor please :-) >>> >>>aloha >>> martin >> >> >>I don't save 'em... >> >>but from experience.. >> >>Crafty vs a GM >> >>at 1 0 will win almost every game >> >>at 5 0 will win most games by a big margin but will draw or lose one here >>and there. >> >>at 15 0 will win the majority of the games but will lose a few more and draw >>several more. >> >>at 30 0 will win more than it loses with even more draws thrown in. >> >>at 60 0 it might "break even" or do slightly better. >> >>The curve is clear. As you give both more time, the humans do better and >>better vs the computer. It has _always_ been this way... IE when Deep Thought >>was producing its 2650 over 25 games, IM Mike Valvo played it two games at one >>move per day and totally trounced it in both, while at blitz he had absolutely >>no chance and at 40/2hr would have had great trouble holding on to his head... >> >>Another way is to check out computer ratings on ICC. Bullet ratings are always >>higher then blitz, while blitz ratings are always higher than standard. Same >>idea... > >hi bob, > >thanks for the description. now, if you could only save these numbers for a few >weeks and post them here, this discussion could be laid to rest :-) > >aloha > martin actually it should be possible to "mine" them from the crafty.pgn file on my ftp server. It has a ridiculous number of games crafty has played on the chess servers since it started playing...
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.