Author: Marc van Hal
Date: 16:50:16 06/25/99
Go up one level in this thread
On June 25, 1999 at 09:37:41, Robert Hyatt wrote: >On June 25, 1999 at 08:24:09, Bernhard Bauer wrote: > >>On June 24, 1999 at 09:50:51, Robert Hyatt wrote: >> >>>On June 24, 1999 at 00:39:39, Sarah Bird wrote: >>> >>>>On June 23, 1999 at 18:22:28, Robert Hyatt wrote: >>>> >>>>>On June 23, 1999 at 00:37:10, Sarah Bird wrote: >>>>> >>>>>>On June 22, 1999 at 20:12:22, Robert Hyatt wrote: >>>>>> >>>>>>>On June 22, 1999 at 15:41:58, Howard Exner wrote: >>>>>>> >>>>>>>>I think different versions of Junior have logged in >>>>>>>>40/2 tournament games against strong humans. Anyone have stats on these >>>>>>>>results? I vaguely recall it doing quite well even on slower hardware. >>>>>>>> >>>>>>>>What is the time control for the upcoming Karpov - Shredder game? >>>>>>>> >>>>>>>>Slowly a collection of tournament condition >>>>>>>>40/2 encounters will put to rest the speculation. >>>>>>> >>>>>>> >>>>>>>Just so we follow formal 'sampling theory' here. IE we do _not_ want to pick >>>>>>>a good result by Junior without picking all the bad results. Easier is to take >>>>>>>these Rebel games and other acceptable games as they are played, rather than >>>>>>>going back. Because to sample backward you have to include _all_ the data >>>>>>>points, else 'cherry-picking' will greatly bias the result... >>>>>> >>>>>>Following excerpt is from IM Larry Kaufman's review of Hiarcs 7 >>>>>>"HIARCS, by Applied Computer Concepts Ltd. with chess engine by British >>>>>>programmer Mark Uniacke, has been one of the very strongest programs for the >>>>>>last several years. The current version, 7.0, is apparently no exception. The >>>>>>latest Swedish rating list (the most widely accepted standard for comparing >>>>>>computer programs) ranks it third, just an insignificant 9 rating points behind >>>>>>the co-leaders (CM 6000 and Fritz 5.32) and substantially ahead of the latest >>>>>>rated versions of such strong programs as Junior, Rebel, MChess Pro, and Genius. >>>>>>Moreover it is up an impressive 43 points from its predecessor, Hiarcs 6. To >>>>>>fully appreciate just how strong Hiarcs 7 is, consider that its Swedish rating >>>>>>of 2567 was earned on hardware (200 MHz MMX) markedly inferior to the latest >>>>>>models (450-500 MHz). Moreover, the Swedish ratings are particularly severe, >>>>>>almost certainly more conservative than FIDE ratings and far below USCF ratings. >>>>>>These ratings are based on 40/2 games with other computers, with the overall >>>>>>level of the list based on games with human competition some years ago. Although >>>>>>I suspect that the level of the top computers may be a bit overstated now due to >>>>>>failure to recalibrate the list based on today's GM level computers, this should >>>>>>be offset by the severity of Swedish ratings in the past, so my guess is that >>>>>>the 2567 rating at 200 MHz would hold up in FIDE competition today, which would >>>>>>imply a FIDE rating over 2600 on today's fast machines. In other words, HIARCS 7 >>>>>>plays tournament chess on a par with the top five players in the U.S. This in >>>>>>turn implies that at action chess (game/30') HIARCS 7 probably plays around 2700 >>>>>>FIDE level, on a par with the number ten player in the world, and should play >>>>>>blitz better than Kasparov, Kramnik, and Anand." >>>>> >>>>> >>>>>Larry is entitled to his opinion. However, he was writing about 2400 programs >>>>>in the days of the 486/33 too. And while he can write about them, it doesn't >>>>>mean that they are there yet. Lets just wait for a while and see whether the >>>>>gap widens (5 to 3 so far) or gets closer, or if the programs can actually pass >>>>>the humans... >>>> >>>>Robert, my own personal opinion is that they are not at that level however I >>>>always respect the rights of others to believe that they are. I honestly feel >>>>that if computers were playing in high rated events say cat 15 and up that GM's >>>>would eat them up after some heavy preparation against playing computers. Which >>>>means if computers were allowed to participate in these events GM's would be >>>>forced into preparation in the same way they now prepare lines for other GM's. >>>>Sarah. >>> >>>You are right of course. The folks that continually talk about programs at >>>GM level don't know any GM players. I've been trying to convince one of several >>>GM players to maybe participate here in a discussion with everyone, because they >>>can quickly explain just why programs are not there yet. And if we could get >>>a program into FIDE, where results are important, preparation would be a >>>_serious_ problem... >>> >>>I think it will be hard to get a GM to discuss this however, because they _know_ >>>that the discussion will turn ugly when someone disagrees with them... >> >>When I play my kind of advanced chess (playing whith a look at the score and >>taking back moves when I think it's appropriate) I usually win. That would not >>happen if I would play a grandmaster. So computers are at present no >>grandmasters. Computers have often no idea about the position. >> >>Another point: If you had a difficult position, would you prefer to give it to >>fritz etc. or would you prefer a grandmaster to have a look at it. >> >>Here is an easy position. How long takes your program to find a win for white? >>FEN: 3k4/q7/3p/p2P1p/P2P1P///K6Q w >> >>Kind regards >>Bernhard >>to analyze it for an hour > > >Just so you know, many a GM can't solve fine 70, which is what this position >is all about. :) I was at the ACM tournament where the chess 4.x guys reported >that they could solve this position (23 plies, 23 minutes of cpu time). At >least one GM present didn't solve it after a couple of minutes... > >Most programs will likely win it. IE Crafty sees check check trade queens and >now we are really at fine 70. On my xeon, it might find the win, as it has to >go 18 plies from the original fine 70 position to see the win. which is 24 >plies from the starting position. It will take a while however... Ofcourse something else have to be said here. Normaly a grandmaster is specialist in one to 5 openings for each side and only try to find the better lines in the specialist openings and it is dificult for a computer to find all the time the best moves in all positions this is the job for the ones who are making the openingbooks actualy if you want to have Hiarcs or Nimzo or any other program with a learning option you should just as the grandmaster only specialize your self in 6 or 6 openings but for instance not the one time e4 and the next time d4 also a grandmaster ofcourse as every human can look stratigal still a lot further then a chessprogram and indead the saying about a position wich a grandmaster did find in a kind of position in 2 minutes and a computer in hours wasn't it also so that jusupov sory no ofence torwards you but overlooking sometimes a mate in 2? Most defenetly programs are tactacly way above a grandmaster that is also why Grandmasters like to change of pieces to go to an endgame look to the games Fritz5 versus Anand So for now it is imposible indeed to win all games against a grandmaster and they even can use Comp comp games against the program Rebel10.5 versus GM Rhode was a perfect example of that Remember that I told you that I had exactly the same game on my harddisk but they players there where Rebel8 versus Chessmaster 6000 if you dont believe this look for your self that is if you have both of these programs
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.