With 6 or 7 pieces left engines have 'perfect play' through tablebases, but when you present the engines with a drawn position the analysis is sometimes unreliable. The engine won’t make an effort to win, because they assume perfect play from the opponent as well. Thus, in many cases that are often won in practical play, the engines may show a dead draw. Also in many cases where material and/or positional factors offset each other, they will show an evaluation of =0.00. In fact, in many of these situations the position may not be drawn at all, but other factors, usually recognized by strong players, may mean that one side is actually lost. And because sometimes engines think the position is drawn, they will make no effort to apply pressure to force errors because they do not know how. Sometimes engines don’t recognize danger until it’s too late and only when they begin to recognize the danger will they start to fight back.
I have an ongoing game right now where several engines said I have nearly a one Pawn advantage but not one of them seemed to suggest a continuation that looked like it would lead to a win; all they were doing was shifting pieces around without any real ‘plan’ that I could see. Because the engines weren’t offering anything constructive, I played a move that was not recommended by any of them and their evaluations actually showed the move to be about one quarter of a P better than their own suggestions. I’m not sure about the win though because it looks like an opposite color B ending is likely. True, shootouts give me a win in about 100 moves. That’s not real reliable but at least it’s my opponent who has all the losing chances.
Houdini 1.5a is free and has risen to the top spot on every rating list that will include it and programmer Robert Houdart has admitted that he incorporated code from previously existing chess engines such as Firebird and Crafty. What he ended up with is an overwhelming strong positional program.
From what I’ve read, Firebird is considered by many to be one of the best computer chess programs in the world. Of course there is also some controversy associated with Firebird as Vasik Rajlich, the founder of Rybka, has alleged that the programmers of Firebird copied nearly all of his code. However, this claim has never been substantiated and after Rajlich and Rybka got busted for ‘code stealing’ who cares?
The real test is how do these engines do in ‘real’correspondence play? For the last 28 games that I have played on LSS I have been using FireBird 1.2v64 and my results are +6 -5 =17. Not too impressive but when I checked that against my previous 28 games where I used Houdini 1.5x64 my results were +3 -7 =18 so the results with FireBird have clearly been better. In addition to switching to FireBird, I also spent some time putting together my own opening book with the result that I began getting better positions out of the openings. I do not have endgame tablebases installed on my computer, but have, as the endgame approaches, gone to the Shredder online endgame database and looked at potential endings. This has saved a couple of games and also allowed me to squeeze out a couple of wins. Still a +1 isn’t anything to crow about, but it’s better than a -4 and I attribute the better results to be the use of FireBird, a better opening book and checking potential endings against the tablebase. It appears breaking the 2200 barrier on LSS is still a ways off unless I can think of something else. Oh, there is one thing...I could study and try to improve.