My Houdini is weird! In the first position it plays 1 ...Qh6, giving it as winning and making me think it's [i]possible[/i] it's gone for this having evaluated it as actually stronger than 1 ...g3 2 h3 Bh3. Then, if I input 1 ...g3 it then sees 2 h3 Bh3 immediately (playing 2 hg in fact), but when I input 2 h3 as well, it plays 2 ...Nc8. Weird! ...

one of my favorite games is still this: http://www.chessgames.com/perl/chessgame?gid=1048284
A game which computers still don't understand the forced win for white. It came up as a 'test your strength' puzzle on chessgames.com Everyone cheated and declared the position drawn by writing down their computer's analysis. While Shirov did  not play the best defense, the game is heavily analyzed by Gelfand in his 300 best games book. All the computer moves are refuted by Gelfand's analysis to wins. No draw!

As correspondence chess players know...the engines are far from perfection. One may look at this position with Houdini, Rybka or whatever and never find the win for black (position taken from a correspondence chess game):


Then try to think yourself, analyze a bit...and you'll see:
1... g3! 2. h3 Bxh3! 3. gxh3 g2 4. Rfc1 Qh4-+ 
They use to to say correspondence chess is dead. Yes, it's dead...when relying only on machines. 

Smyslov_Fan wrote on 11/02/11 at 12:42:42:


Can you link to that article?


BTW, I also concur with Markovich and MNB. Chess engines are not god and they are not telling 'Chess Truths'

About engines, often different engines tells completly different things in complicated positions.
I remember a certain sharp Dragon line where first engine says white is winning and second engine says black is slightly better and the truth was the black was much better.

I can understand why the position in question is eventually equal. Without calculating a lot of variations 
For White to win he must prevent that Black exchanges his pawn and 2 pieces against the 3 pawns (2 lonely knights cant deliver mate). That may prove extremely difficult, especially since the 'defenders' king is in front of the passed pawn(s).
But there are plenty of other endgame positions that are too difficult for engines at this moment (perhaps not in the future). Another example are positions with R+p vs. B+Kn, Rook end-games with multiple pawns (4p vs 3p) are often too difficult for a correct evaluation.
I also appreciate the comments on gambits. Especially those gambits where compensation for a pawn in a semi-closed position is in the nature of a creeping initiative
where you can slowly improve your position further and the opponent has been robbed of any pawn-brakes.

Centaur >> Engine, of course. I  agree with it.

@Smyslov_Fan,

I know there are some positions which engines can't understand in a reasonable time, or never. I have a very sweet position like this in my mind too. Fact is this position is also a sharp tactical position, there is a forced mate, and engine couldn't solve it! I have solved it with playing pieces in half hour on my chessboard. Meanwhile Rybka was pondering the solution but she couldn't find until i finish my solution.

I have analyzed your position with my engine, only 4 million nodes and the result is 0.00. I agree with Houdini, white can't win. But i would prefer white here, as whites play easier than blacks. Here is the analysis:

Analysis by Houdini 2.0 Pro x64:

1.Nf6+ Kc7 2.c4 Rh4 3.Ng4 Bg6+ 4.Ne4 Bf5 5.Ne5 Be6 6.Nc5 Bc8 7.Ne4 Rh5 8.Kc3 Be6 9.Nc5 Bc8 10.Ne4 Be6
=  (0.00)    Depth: 32/20   00:06:38  3878mN


And i waited some more:

Analysis by Houdini 2.0 Pro x64:

1.Nf6+ Kc7 2.c4 Rh4 3.Ng4 Bg6+ 4.Ne4 Bf5 5.Ne5 Be6 6.Nc5 Bc8 7.Ne4 Rh5 8.Kc3 Be6 9.Nc5 Bg8 10.Ne4 Rf5 11.Kb4 Bf7 12.Kc5 Bh5 13.Nd6 Rf6 14.Ne4 Rf5 15.Nd6
=  (0.00)    Depth: 34/54   00:22:50  11685mN

Again the same, with deeper line.



I tried to setup my position but couldn't remember completely for now. Maybe later.

I completely agree with Markovich and MNb.

Piling on a bit, but for a good cause: There are some websites where it seems that computer analysis is treated as heaven-sent wisdom. Thank Heavens this isn't one of them!

Experienced correspondence players love these gambits precisely because the computers mis-evaluate them on a regular basis. My feeling is that this may be one of the weakest spots in computer analysis, along with material imbalances such as Rook+B vs 2 Knights and 2 pawns. Sure, computers have made huge improvements over the years, but the savvy analyst who uses engines and his own experience will almost always trump the pure engine user.

Here's a random example of a position that I would evaluate as very good for white but the Rybka/Firebird/Houdini family would consider to be slightly in Black's favor:

[