Author: Robert Hyatt
Date: 20:30:10 12/14/03
Go up one level in this thread
On December 14, 2003 at 20:22:50, Omid David Tabibi wrote: >On December 14, 2003 at 05:35:56, Amir Ban wrote: > >>On December 14, 2003 at 04:10:02, Sandro Necchi wrote: >> >>>On December 13, 2003 at 17:49:47, Amir Ban wrote: >>> >>>>On December 13, 2003 at 07:32:04, Sandro Necchi wrote: >>>> >>>>>On December 13, 2003 at 05:24:46, Uri Blass wrote: >>>>> >>>>>>On December 13, 2003 at 03:32:01, Sandro Necchi wrote: >>>>>> >>>>>>>On December 12, 2003 at 16:59:17, Anthony Cozzie wrote: >>>>>>> >>>>>>>>>My point is: >>>>>>>>> >>>>>>>>>1. Since the programs now are much stronger than 20 years ago, why not change >>>>>>>>>the rule about resigning and let them resing when they are down -10? >>>>>>>>>2. It is true that a bug may help the program which is lost, but which are the >>>>>>>>>chances today? Is it correct to say 1 every 1000? If this is true, why not >>>>>>>>>concentrate to improve their play on the first part of the game rather then >>>>>>>>>hoping to be extremely lucky in the endgame? >>>>>>>> >>>>>>> >>>>>>>Hi, >>>>>>> >>>>>>>first of all thank for the friendly discussion. I undestand your point of view >>>>>>>and I do respect it as I do with everybody points of view. >>>>>>>Still I do not agree with you...see below. >>>>>>> >>>>>>>>The point is, even if the eval is -10, I am under no obligation to resign. >>>>>>> >>>>>>>Correct. >>>>>>>I am asking to change the rule to force a program to resign when the score goes >>>>>>>down to -10 (a mean more or less a queen and 2 rooks down, to summarize). >>>>>> >>>>>>The problem is that with the new rule programmers have no problem to change >>>>>>their evaluation and never show a score of more than -9.999 pawns against >>>>>>themselves even in case of mate. >>>>> >>>>>Ok, this should be verified with a secret position before the tournament start. >>>>>If a programmer is found as cheating, than unless he can demostrate it is due to >>>>>a bug for that specific position it will be disqualified. >>>>> >>>>>> >>>>>>It is easy to do it for me by dividing all scores by 10 so 99.99 that is mate in >>>>>>one today becomes 9.999 >>>>>> >>>>>>The only way to implement it correctly is if an external program does the >>>>>>evaluation. >>>>> >>>>>Maybe. Mine is a proposal. Maybe there is a better idea to handle this. >>> >>>Hi, >>>>> >>>> >>>>In WCCC 2001 Shredder lost a game to Junior where it was showing +6 at some >>>>point. >>> >>>I know it very well. >>> >>>> >>>>The rule at WCCC is that the TD must allow resignation or agreed draws. E.g. the >>>>draw Junior-Fritz was already agreed several moves back, but the TD wanted to >>>>see the rooks off before giving consent. I think that Jonny was not allowed to >>>>resign, and rightly so. In the critical phase of the game black had mate threats >>>>of its own, and with the sort of bugs Shredder was suffering, even a loss was >>>>not inconceivable. >>>> >>>>It was noted by several after that game that given that operators are not >>>>allowed to resign or draw when they want to, it's ridiculous to allow them to >>>>lose deliberately. >>> >>>Amir that is all close and past now. >>>I am making proposals for the next tournament to make thinks more clear to >>>everybody. I think that what turned out did show that things were not fully >>>clear as there were too many different opinions. I do not agree with your >>>opinion, but I respect it. >>> >>>My proposal is: >>> >>>1. Let's change the rule (when the tournaments have live games; when there are >>>people looking them other than those in the hall) to force all programs to >>>resign when they are at -10 or lower. >>>2. Since the chances that these games turned our to be a draw or change the >>>outcome are nearly 0, let's avoid showing parts of games which are not played in >>>human tournaments and not interested from a chess point of view. I think -10 is >>>a good value to avoid maybe 1 game out of 1000 to change the outcome. >>> >>>Of course without cheating... >>> >>>P.N. The bug in Shredder has been fixed since several days, so my proposal has >>>nothing to do with Shredder. >>> >>>The reason of the proposal is that people watching these games are continuosly >>>saying why program x is not resigning and switching to other games as that part >>>of the game is not important at all. >>> >> >>I agree with this part. The problem however is the reverse: the programs are >>just too damn smart for ordinary people. They will say -8 based on a deep >>continuation and resign, leaving many viewers mystified. Homework for you boys >>and girls: why did program X resign ? How many can see the mate in >>Shredder-Jonny ? Or, viewing the position and seeing that black has mate-in-one >>threats, even conclude that white is winning ? The first report I got after the >>reptition was from a kibitzer who said excitedly: "Shredder is now losing", >>which turned out to be a gross exaggeration. The TD's are mindful of this and >>want to see games played until the t's are crossed and i's dotted. >> > >A personal example: > >In the game ParSOS - Falcon, I resigned in the following position: > >[D]8/5k2/5n1P/1R6/6p1/4Kp2/8/8 w - - 0 86 That is a semi-reasonable position to resign in. I get +9 in about 30 seconds and depth 16. Depth=18 comes at 60 seconds and +10.5 However, it would be _much_ better for your program to resign than for you to do it. My preference is for the operator to be _passive_. That's what led to the debacle in Graz in the first place. > >It is very hard for a human to see why white is winning. Just as the game >finished I logged in to PlayChess server. The result was still not appearing >there, and I saw many people discussing about whether white or black is better >in that position :) > >For the sake of observers of course I should have continued playing. But after 5 >hours (of a boring game), I was too tired to continue just for the sake of the >observers (especially since 6 moves earlier my program blundered a move, losing >the drawn game...). > > >>Amir >> >> >>>This is just a proposal. >>>If you programmers do not agree, then forget it. >>> >>>P.N. I like the way your program play. I find it fun! >>> >>>>Amir >>> >>>Sandro
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.