I'm coming up for air after doing the daily round reports and analysis on Corus Wijk aan Zee for ChessBase.com for the past few weeks. Argh. So much analysis only to show a a couple of diagrams and lines, but you don't want to miss any of the decisive moments. That's always embarrassing. Is this where I can yet again tell my story about showing a win that Short and Radjabov missed in analysis of Radjabov-Anand, Linares 2003? It was in my report the same day of the game, for goodness sake. A few days later Short gives an erroneous draw in his Sunday Telegraph column. A few weeks later Radjabov himself gives the same nonexistent draw in New In Chess. Months later GM Krasenkow writes in to New In Chess to say some students of his found 'this amazing win for Black" blah blah. Same thing I gave the day of the blooming game. But I'm not bitter.
I think even GMs are so dependent on computers these days that many of them don't really know how to use them for analysis. They scroll through the moves expecting a few second to illuminate the secrets. This isn't true even in very tactical positions. I find out more things because I often use Fritz and Co. to explain to me why my ideas don't work. On rare occasions they DO work and it's something the machine would have needed a long time to find on its own. After I use Fritz to auto-check games for blunders I always have to spend a few minutes going through the score to remove some of the silly evaluations it gives. Trust them for most tactics, and you have to love endgame tablebases, but it's not going to revolutionize chess looking each move of a super-GM game for five seconds.
This is what drives me crazy about kibitzing live games at Playchess.com, and it's no better on other chess servers. Having 800 amateurs shouting about how Anand and Kramnik are "blundering" because their moves are rated -0.34 worse than what their chess engine wants to play. This is a joke. Unless the eval drop is more than a full pawn (1.00) after a good five minutes of computer time, skip it!
Here's an experiment you can try that proves my point. Take a handful of super-GM classical games. Have Fritz auto-annotated them with the analysis threshold at 60, so it will only add variations if it thinks its suggestion is that much better (over half a pawn) than what was played in the game. Give it five or ten seconds per move. When it's done, have it do the same games again but with one or two minutes per move, or even more if you can let it run overnight.
The result? The more time you give it the LESS analysis it adds. That is, Fritz comes around to agreeing with the GMs instead of finding more so-called mistakes. Viva la humanidad!
