Author: Robert Hyatt
Date: 17:06:51 04/26/98
Go up one level in this thread
On April 26, 1998 at 11:50:01, Don Dailey wrote: >>> Hi Bob, >>> >>>Again, as I mentioned in a previous post, I think you over-rate the >>>time control advantage for computers. I do not deny that computers >>>are better relative to humans at faster time controls, this is well >>>known. But 10-10 is starting to get pretty comfortable for human >>>players. >>> >>>If a computer is indeed in the same league as some given player at >>>10-10, you should not expect to be completely outclassed at tournament >>>time controls. The humans will certainly play MUCH stronger at 40/2 but >>>so will Crafty. In my opinion the difference is not as HUGE as you >>>would have us believe. >>> >>>- Don >> >> >> >>I think it's bigger than you think. IE I personally would expect to >>lose >>at least 3 of every 4 games at 40/2 vs a GM like Larry or Roman. Yet >>both >>are maybe 2600 level... when I lose 3 of 4 (which means I win 1 of 4 >>which >>is not a bad results when you think about it) guess where my rating ends >>up? >> >>2400, that's where... and that's 150 points below what it takes to be >>called a GM... > >Hi Bob, > >You could be closer to the truth than I am. Even though your >numbers seem high to me I don't have any really serious data >to back me up. I have lot's of anecdotes but I'm sure you do >too. > >It would help if someone had a LOT of rated tournament games and >also a lot of fast (but not speed chess) games for the very same >program. I'll bet the USCF has some machines certified at more >than 1 time control. Do they still do these certifications? > >The only problem even with this, is that each rating could easily >be off 50 points or more. That would make a lot of difference >if they were both off in opposite directions! > >- Don You have to take those rating numbers like this: they are based on Elo-like calculations, but with an unknown rating pool size of unknown ratings of unknown accuracy. IE, No I don't think Crafty is 2600 at 40/2... that's silly. But let's assume that ICC's ratings are inflated by "X". Then crafty's bullet rating should be 3100-X, it's blitz rating 2900-X, and its standard rating 2600-X. Now, whatever X is, it changes nothing about comparing crafty at blitz vs IM/GM players, vs crafty at standard vs IM/GM players. It clearly plays 300 points worse at standard than at blitz. And this is pretty well the norm for *all* programs that play on the servers... the faster the time control, the higher their rating. In fact, we once had a program called "hill" on ICC that only played 1 0 and it had a *huge* rating on fairly slow hardware, solely because at 1 minute total time, humans simply can't cope...
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.