Author: martin fierz
Date: 23:09:11 08/28/04
Go up one level in this thread
On August 29, 2004 at 01:45:53, Kurt Utzinger wrote: >On August 28, 2004 at 18:40:43, martin fierz wrote: > >>aloha! >> >>after all the cloning talk, here's something to test your favorite engine with: >>i played a weakish IM today and got the following position after 56...Qg6+ >>57.Kf2: >> >>[D] 8/6kp/8/3p4/1P4q1/P4p2/5K2/4Q3 b - - 0 57 >> >>it took me a couple of seconds to see that after 57...Qh4+ 58. Kf1 Qxe1+ 59. >>Kxe1 Kf6 black is completely winning due to the floating square (it's nice to >>see that reading dvoretzky is some good!) and so i played that and won :-) >> >>interestingly, when i entered this game in chessbase running fritz7 in the >>background, fritz didn't seem to see this at all within a minute or so. so >>question to all: how long does your favorite engine need to see the line ...Qh4+ >>Kf1 Qxe1+ with a HUGE plus score (i.e. +9 or so - that is what my assessment of >>the position was after a couple of seconds...)? >> >>i don't want to claim that this is the only win, but it is by far the easiest >>one, and fritz 7 misses it completely. >> >>cheers >> martin > > > Hi Martin > A most interesting position showing the difference > of thinking between human and machine. You "know" > that the exchange of queens is an easy win, the > engines have no idea. hi kurt, yes and no. for example, crafty knows something about split vs connected passers and unlike hiarcs in its analysis you posted below would play 60...Qe4+ and exchange queens for a winning pawn endgame. crafty also first gave a line with Q-exchange which it wrongly evaluated as winning for black, when in fact it was a win for white. my fritz 7 seems to be more on the agnostic side, even after you enter the pawn ending it doesn't give high scores for a couple of seconds up to ply 17 or something like that. so this position shows a bit whether an engine has this type of pawn-endgame transition knowledge or not, it also shows what the engine does about transitions to pawn endings. this transition has been a subject here a couple of times; some engines are better at it than others (i would very much like to see what yace makes of this - dieter?) > But there is one important > thing to consider here: for computers 57...Qg2 > wins too and this line may be the reason that > the engines can't find the more human like 57...Qh4 > 58.Kf2 Qxe1 within reasonable time. that is probably true, but it would be easy to modify this position so that only the Q-exchange wins - then only engines which manage such transitions would solve it. for the record, my own program (which has some code to help it for such transitions) fails miserably on this test of course! cheers martin
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.