Author: Robert Hyatt
Date: 16:01:03 11/25/00
Go up one level in this thread
On November 25, 2000 at 16:21:58, James T. Walker wrote: >On November 25, 2000 at 13:35:41, Robert Hyatt wrote: > >>On November 25, 2000 at 13:06:43, James T. Walker wrote: >> >>>On November 25, 2000 at 10:00:18, Robert Hyatt wrote: >>> >>>>On November 25, 2000 at 09:47:30, Lin Harper wrote: >>>> >>>>> Just now I was watching GM Kotronias on ICC, playing Rebel Tiger 13. >>>>> 3min blitz, he got one draw! More than I'd do I suppose, but is this >>>>> what's happening now? GM's hopelessly outclassed at quick chess? >>>> >>>> >>>>At 3 mins/game, this is not uncommon. at 5 3 it is more uncommon. at longer >>>>time controls it generally doesn't happen at all. >>> >>>Hello Bob, >>>I assume you mean that at longer time controls the GM's are not hopelessly >>>outclassed at all. That's probably true but I would remind you of the drubbing >>>the GM's & IM's took at the hands of Deep Fritz & Deep Junior at KC not very >>>long ago. That was at Game/1hour! So this begs the question of how much time >>>do GM's need now to play _e_v_e_n_ with the machines? >>>Jim >> >> >>Bring any program you want to ICC, and let me get a couple of GMs to play >>it several dozen games a day for a week. And watch what happens. (at a >>time control of say 5 10). They will do badly at first, but they will >>begin to "solve" the problems presented by the programs. No the humans >>won't win all games. They might not even win the majority. But there will >>be _no_ 14 - 0 - 1 type results. >> >>I have played Roman so many games, that he knows _every_ weakness Crafty has. >>Fortunately he (and a couple of others) will report these weaknesses so that I >>can address them. But he is a _serious_ problem for computer chess programs, >>even though he really doesn't play as much as he did 20 years ago (in human >>tournaments). >> >>Some seem to think blitz is "solved". It isn't, yet. At 3 0? Computers >>are tough but they will _not_ win every game. Nor even every 4 game match. >>Which is amazing, IMHO. To find out how good your favorite program really >>is, put it on ICC and _leave_ it there for a week or two. Letting humans >>play as many consecutive games as they want. The results will be interesting. > >Hello Bob, >Yes, I think it would be interesting. Wish I could do that. One thing I don't >understand is why you are not more interested in playing Crafty vs other >computers. It seems to me that other programs have the ability to point out >many other weak points in Crafty that humans may not. My point is, who cares? Computers don't attack worth a flip. Humans do. If you write a crappy king safety evaluator, and play against computers, you might not even notice. Computers don't make the same sorts of tactical mistakes that humans do. Crafty will lose many games because it is simply "slow". It would be fairly easy to make it 2x faster. But that would make it 10x harder to modify in the future. I choose to stay "slow" for that very reason. And I _know_ I am slow... And it doesn't take tactical losses against computers to let me know... I am interested in the evaluation issues, not the search/speed issue at present, as hardware will slowly erase search problems... Crafty is playing almost nothing _but_ computers at present, which is what I was afraid would eventually happen. > Maybe Crafty would be a >more well rounded program if you played more computers and analyzed it's losses. > This is not a criticism but only an idea of mine. When I had a computer >account I played Crafty occasionally and won as many as I lost even though I was >using slower hardware. Can you analyze these types of games and see any >weakness that doesn't show up vs humans? Maybe it's less time consuming for you >to let the GM's point out the weakness in Crafty and spend more time working on >those weaknesses. I was thinking that even if one loss to a computer does not >show an obvious weakness that several may start to show a pattern. >Regards, >Jim
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.