Author: Robert Hyatt
Date: 08:19:23 10/24/02
Go up one level in this thread
On October 24, 2002 at 03:06:40, martin fierz wrote:
>On October 23, 2002 at 23:56:39, Robert Hyatt wrote:
>
>>On October 23, 2002 at 23:33:43, martin fierz wrote:
>>
>>>On October 23, 2002 at 20:38:57, Robert Hyatt wrote:
>>>
>>>>On October 23, 2002 at 19:00:15, martin fierz wrote:
>>>>
>>>>>On October 23, 2002 at 15:16:39, Robert Hyatt wrote:
>>>>>
>>>>>>On October 23, 2002 at 14:54:09, martin fierz wrote:
>>>>>>
>>>>>>>On October 23, 2002 at 11:26:38, Robert Hyatt wrote:
>>>>>>>
>>>>>>>>On October 23, 2002 at 05:08:11, Vincent Diepeveen wrote:
>>>>>>>>
>>>>>>>>>On October 22, 2002 at 17:29:53, martin fierz wrote:
>>>>>>>>>
>>>>>>>>>let's be clear. the kramnik guy was happy to receive
>>>>>>>>>a million dollar in advance. Without much effort he played
>>>>>>>>>a few moves and it was 3-1. Then everyone started complaining
>>>>>>>>>that the match got no publicity and got no excitement.
>>>>>>>>>
>>>>>>>>>He then gives away a piece in a clear drawn position with
>>>>>>>>>a 1b trick (1 check in between). That's bullet blunder level.
>>>>>>>>>
>>>>>>>>>In fact i don't make such mistakes that much at bullet and
>>>>>>>>>last time i made such a mistake at slow level was a year or
>>>>>>>>>10 ago. Kramnik had plenty of time.
>>>>>>>>>
>>>>>>>>>0% chance he didn't deliberately blunder there.
>>>>>>>>
>>>>>>>>I think that is a totally stupid statement to make. I can point out GM blunders
>>>>>>>>in _every_ tournament I have watched online. I have seen them overlook a mate
>>>>>>>>in 2. A hanging queen. You-name-it. Human GMs _do_ make mistakes. Not as
>>>>>>>>often as non-GM players, but also far more often than "never".
>>>>>>>
>>>>>>>bob, of course human players make mistakes. but GM != GM. kramnik is way beyond
>>>>>>>your average GM. i challenge you to find a tournament game ("normal" time
>>>>>>>control, not rapid chess) by kramnik in the last 5 years where he made such a
>>>>>>>blunder without time trouble. i'd be surprised if you found one :-)
>>>>>>>(but i'd really like to know the answer to that one!)
>>>>>>
>>>>>>I don't have a large database of games to look over, so I am really not sure.
>>>>>>My
>>>>>>observation was based on actual live games being relayed from major human
>>>>>>tournaments
>>>>>>on ICC, where Crafty was giving online analysis to make spotting the blunders
>>>>>>much easier.
>>>>>>
>>>>>>I saw one game where white hung a piece, black didn't notice, and on the next
>>>>>>move white
>>>>>>"corrected" things and the game continued. Had black took it was an instant
>>>>>>loss for white.
>>>>>>In another game, white made a move that forced him to give up a queen the next
>>>>>>move or
>>>>>>get mated in 2 moves. Very simple blunder. Both were 2650+ players at the
>>>>>>time. I think
>>>>>>one might have been Leko but I am not sure... This is not nearly as uncommon as
>>>>>>it seems,
>>>>>>and many blunders go unnoticed by the opponent, making them "silent blunders"
>>>>>>that don't
>>>>>>get noticed by anybody...
>>>>>
>>>>>please note that i am talking about kramnik, and about non-time-trouble. i
>>>>>remember very clearly when karpov lost a piece against christiansen on move 10
>>>>>with a simple check i think - very like kramnik now. why do i remember this?
>>>>>because this position was published in *every* single chess magazine of the
>>>>>world, saying: "look, karpov is only human too".
>>>>>there is a HUGE difference between kasparov's blunder you qote (resigning a
>>>>>drawn position) and the blunder kramnik made. i know that you are *by far* good
>>>>>enough at chess to see that the difficulty level of these two blunders is miles
>>>>>apart. one is a simple 3-ply search. the other is, as you wrote recently, a day
>>>>>or so of analysis by a bunch of CCC members and their machines.
>>>>>if kramnik had made a blunder of this magnitude in the last 5 years in a
>>>>>tournament game, i'm pretty sure it would have been all over the chess magazines
>>>>>and i would have seen it...
>>>>>
>>>>>the worst blunder in a world championship match i can remember is a bad rook
>>>>>move (...Re8 or something like that) by karpov in one of his matches against
>>>>>kasparov, which lost "on the spot", but that was much more than a 3-ply search,
>>>>>and combined those two have probably played about 100 games.
>>>>>
>>>>>>So,
>>>>>>IMHO, it just
>>>>>>goes down as "yet another GM blunder, which _does_ happen from time to time."
>>>>>of course this is quite possible. but you can look at say kramnik's last 500
>>>>>classic tournament games and look how many times he blundered a piece that a
>>>>>3-ply search would find. all i'm saying is that the fritz team hit the jackpot,
>>>>>because normally kramnik would not make that kind of blunder even in an
>>>>>80-game-match...
>>>>>
>>>>>aloha
>>>>> martin
>>>>
>>>>
>>>>I think there are two issues:
>>>>
>>>>(1) I don't know what the probability is that he would make a major blunder in
>>>>an 8-game
>>>>span. Probably very low. So, once again, serendipity strikes, this time in
>>>>favor of the computer,
>>>>where it often strikes in favor of the human. :)
>>>
>>>yes, i guess that is true...
>>>
>>>>(2) It is more than possible that some of his mistakes have gone unnoticed,
>>>>since I doubt many
>>>>play over every game of his using a computer.
>>>i doubt that "many" do this, but one person per game is enough - to find 3-ply
>>>losers!
>>>
>>>
>>>> But it would be interesting to
>>>>get a file of (say)
>>>>his last 100 games and sic a computer on them in "annotate" mode to see if it
>>>>finds anything
>>>>of interest... I have the computers to do this if someone has a set of games to
>>>>check out...
>>>
>>>i can get you 100 kramnik games if you like - i "only" have the big2000 database
>>>of chessbase, but i can get you the 100 last kramnik games in there that and
>>>send them to you by email - does that sound ok? i'll just take the last 100 (and
>>>remove anything that looks like blitz or rapid), other than that i will not
>>>select anything, so as not to bias the outcome of the experiment.
>>>
>>>aloha
>>> martin
>>
>>
>>Fine by me.
>>
>>For some specifics. I will run these on quad 550's using 4 processors. What
>>would be a
>>decent time per move to annotate the games? IE I probably shouldn't search too
>>long if we
>>are looking for relatively simple blunders.
>>
>>Maybe we should define "relatively simple blunders" in some precise way? IE I
>>need two
>>things: (1) time per move; (2) threshold between the move played and the best
>>move according
>>to the computer to trigger an "aha!" comment...
>>
>>suggestions???
>
>hi bob,
>
>i've sent you the pgn with 184 kramnik games from 1997-1999 - he already had a
>2750 rating back then. my suggestion would be 1s/move and 1.5 pawns. i don't
>think more time is necessary than 1 second, since we are talking about 3-ply
>blunders - actually 1ms would suffice for that :-)\
OK. I'll run that and put the annotated files on my ftp machine. That won't
take very
long if we really go for one second. Of course, there will be more work to do
as we will
have to take every "blunder" and re-check it to deeper depth, as I have seen
plenty of cases
where a GM plays a move while crafty is "watching" and it is quickly convinced
that move
is much worse, but as the depth increases, so does the score, eventually proving
that the two
moves are likely transpositions of the same line.
>but i'd also be interested if there were a bit less obvious big errors, so 1
>second should be fine. it depends a bit on how long you want to bog your machine
>down with this - 40 moves * 200 games = 8000 positions to check, with 10s/move
>it would take a full day. the threshold would have to be at least a pawn, but
>not 3, as you can usually still get a pawn for a piece if you blunder it.
>
>once you have run the test we'll have to look at all positions which crafty
>pinpoints to see whether e.g. kramnik just missed a mate in 7 but won all the
>same with a more human approach, or to see if he sacced something somewhere or
>whatever. i don't expect too many positions to come up from these games, so we
>should be able to check them manually - but you never know, perhaps i'm in for a
>surprise :-)
>
>aloha
> martin
>
>>
>>Meanwhile send me the PGN and I will get things ready...
This page took 0.01 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.