< Earlier Kibitzing · PAGE 9 OF 9 ·
|Mar-14-14|| ||Cheapo by the Dozen: Continuing, if Black declines the initial sacrifice and does something to prevent Qh6+, then White is ahead a pawn and still has an attack. But otherwise, I think the line I had before is substantially forced, namely:|
33 f5 hxg4
34 Qh6+ Kg1
35 fxg6 Bxg6
36 Rg5 Kf7
37 Rf1+ Kxe7
38 Rxg6 (or Qg7+)
and Black surely gets shredded by White's major pieces.
Anyhow, I'd play that over the board, and figure out Move 38 when we got that far.
|Mar-14-14|| ||Cheapo by the Dozen: Not that I have the endgame technique to win a won position like that, but I'll say I solved the puzzle anyway. :)|
|Mar-14-14|| ||MostlyWatch: A game between two computers is a game between the two programmers. So a game of man v. machine is a game of man v. the computer programmer.|
For the computer programmer a chess game is a problem searching the tree of possible moves. The search, rather than looking for the best move, could be to look for an avenue with a surprise down the line, or something else. And of course the main endeavor is finding a way to see farther over the horizon than the opponent. The computer programmer can approach the problem in a lot of different ways - the human element is simply moved to a different location, into the algorithm.
|Mar-14-14|| ||al wazir: White is a ♙ up at move 33. After move 38 he is still a ♙ up. All this "combination" does is trade material.|
|Mar-14-14|| ||morfishine: Who could forget this from Last Year's Candidates: Gelfand's e-pawn rolls blissfully to the 7th rank, Aronian scrambles to equalize, then Gelfand finishes him off with <33.f5>|
|Mar-14-14|| ||Cheapo by the Dozen: <al wazir>,
Yes, but without something clever, White would have to immediately give the pawn back, as he had both a piece and pawn en prise.
|Mar-14-14|| ||fishcat: 36.Qg5 is a bit better than trading off the queens.|
|Mar-14-14|| ||Penguincw: I remembered seeing that move before, but didn't get it today. :||
|Mar-14-14|| ||PJs Studio: Fishcat, I thought Qg5 was better too. But when I saw Qe5 I figured I had miscalculated...|
Cheapo - I agree with your initial assessment. Although, I haven't checked the lines against my computer.
|Mar-14-14|| ||Patriot: Material is even. Black threatens 33...hxg4 and 33...Qxf4.|
This proved very tough for me.
33.Bd7 Qxd7 34.Rxh5+ gxh5 35.Qe5+ Kg8 36.Qg5+ Kh7 37.Qf6 Kg8 38.Re5 Rd5 looks very suspicious to me. I think white will be struggling for a draw here.
33.f5 hxg4 34.Qh6+ Kg8 35.fxg6 Bxg6 36.Rg5 Kf7 37.Rf1+
There are lots of other possibilities here but I've already spent too much time on this to be practical.
|Mar-14-14|| ||gofer: We have a Q+2R+B, but it is <Pf4> that is the hero in this position.|
<33 f5 ...>
33 ... Rxg4?
33 ... hxg4
34 fxg6! ...
34 ... Qxg6
35 Qxd4+ Qg7
34 ... Bxg6
37 Rg5 Kf7
38 Rxg6! Qxg6
33 ... gxf4
<33 ... Qf4>
<34 fxg6 ...>
click for larger view
Black's defenses are crumbling its not over yet, but holding on to Ph5 is
looking to be impossible! Which will leave White with two connected passed
pawns. I am guessing from here on.
34 ... Bd5?
34 ... Bg8
35 Qxf4 Rxf4
<34 ... Bxg6>
<35 Bxh4 ...>
35 ... Kg7?
36 Rg5 Qxd3+
37 Rxd3 Rd6
38 g4 Kh7
39 Bxg6+ Rxg6
40 Rxg6 Kxg6
<35 ... Bxh4>
<36 Rxh4+ ...>
36 ... Kh8
<37 ... Kg7>
<38 Qe6 >
Well, I got everything except the best defense. So perhaps a "big"
half a point today! But not much more...
|Mar-14-14|| ||Makavelli II: I found this puzzle very difficult. Surely one cannot be expected to guess from move 33 onwards as this is a 60 move game going on here :) I did try and I did fail :)|
|Mar-14-14|| ||agb2002: White is a pawn up.
Black threatens 33... hxg4 and ... Q(R)xf4.
The airy position of the black king invites to play 33.f5:
A) 33... hxg4 34.Qh6+ Kg8 35.fxg6
A.1) 35... Bxg6 36.Rg5 Kf7 37.Rxg6 Qxg6 (due to 38.Qh7# and 38.Rxd6) 38.Rf1+ wins the queen for a rook.
A.2) 35... Qxg6 36.Rg5 Rd6 37.hxg4 + - [Q+2P vs R+B].
A.3) 35... <any other bishop move> 36.Qh7#.
B) 33... gxf5 34.Rxf5
B.1) 34... hxg4 35.Rxf7 (threatens 36.Rf8+ Rxf8 37.exf8=Q+ Qxf8 38.Qxd4 and 36.Qg5)
B.1.a) 35... Rd5 36.Qc3+ Rd4 (36... Kg8 37.Qg7#) 37.Rf8+ Rxf8 38.exf8=Q+ Qxf8 39.Qxd4+ Kh7 40.Qxg4 + - [R+P].
B.1.b) 35... Kg8 36.Rf8+ Rxf8 37.Qg5+
B.1.b.i) 37...Kf7 38.Qf5+ Kg7 (38... Qf6 39.exf8=Q+ Kxf6 40.Qxf6+ Kg8 41.Re8+ Kh7 42.Rh8#; 38... Ke8 39.Qxf8+ Kd7 40.Qd8#) 39.Qxf8+ and 40.e8=Q wins.
B.1.b.ii) 37... Kh7 38.exf8=Q Qxf8 39.Re7+ wins the queen for the rook or delivers mate.
B.1.b.iii) 37... Kh8 38.exf8=Q Qxf8 39.Qe5+ Qg7 40.Qh5+ Qh7 (40... Kg8 41.Re8+ Qf8 42.Qg6+ Kh8 43.Rxf8#) 41.Re8+ Kg7 42.Re7+ wins.
B.2) 34... Kg7 35.Rxf7+ Kxf7 (35... Kg6 36.Bxh5+ Kxh5 37.Rh7+ Kg6 38.Qh6+ Kf5 39.Rf7+ Qf6 40.Q(R)xf6#) 36.Bxh5+ followed by 37.Bxe8 + - [B+P].
B.3) 34... Kg8 35.Rxf7 hxg4 (35... Kxf7 transposes to B.2) 36.Rf8+ transposes to B.1.b.
B.4) 34... Bc4 35.Bxh5 is crushing.
C) 33... Kg7 34.fxg6
C.1) 34... gxh5 35.gxf7 Kxf7 36.Rf1+ Kg7 (36... Kg8 37.Qg5+ and mate soon) 37.Qg5+ Qg6 (37... Kh7 38.Qh5+ Qh6 39.Rf7+, etc.) 38.Qe5+ wins the rook on d4.
C.2) 34... Bxg6 35.Bxh5 wins another pawn while keeping the attack (White threatens 36.Bxe6, fatally exposing the black king).
D) 33... Qf4 34.fxg6 Bxg6 35.Bxh5 + - [2P]. For example, 35... Qxe3 36.Rxe3 Bxh5 37.Rxh5+ Kg7 38.Rf5 with a won ending.
|Mar-14-14|| ||ajk68: Saw the line after move 33 but wasn't convinced I had the solution. I was looking for something more conclusive.|
|Mar-14-14|| ||rudiment: Saw some elements but my, what a mesmerising line. Complete control from White.|
|Mar-14-14|| ||AylerKupp: <<MostlyWatch> So a game of man v. machine is a game of man v. the computer programmer.>|
Except that the programmer has a superior tool in terms of making calculations. So a better analogy is that a game of man vs. machine is a game of a man with a knife against a man with a gun. Oh sure, the man with the knife has some advantages but I wouldn't bet against the man with the gun.
An interesting idea is to look for a surprise down the line. But how do you define a “surprise”? One possibility is to be able to find a good move that the opponent, human or machine, does not see. And one way to do this is to be able search to a deeper depth than the opponent, but I don’t see how this is any different than seeing farther over the horizon than the opponent.
|Mar-14-14|| ||Patriot: <<AylerKupp>: <<MostlyWatch> So a game of man v. machine is a game of man v. the computer programmer.>> Man created the automobile also, but I wouldn't run a foot race against it.|
|Mar-14-14|| ||Jimfromprovidence: 36 Qg5 looks like an improvement over the text.
click for larger view
The general threat is to get the rook at c5 to e5, then e6.
If black for example, tries 36...Rd5, then after 37 Rxd5 cxd5 38 Bxg6 Qxg6 39 Qxd5 white has a nice advantage.
click for larger view
|Mar-14-14|| ||TheBish: Gelfand vs Aronian, 2013|
White to play (33.?) "Difficult"
White is up a pawn, but it appears that after the attacked bishop retreats, Black will regain the pawn by capturing on f4. But instead of retreating, White can win by opening a path to Black's king.
Now after 33...hxg4 34. Qh6+ Kg8 35. fxg6 Bxg6 (35...Qxg6 36. Rg5) 36. Be6+ Bf7 37. Rg5+ Qg6 38. Rxg6# finishes things quickly.
33...gxf5 34. Rxf5!
Better than 34. Bxf5 Rd5! when one of White's key attackers (the Rc5) is neutralized. Now losing is 34...Bg6? 35. Qh6+ Kg8 36. Rf8+ Rxf8 37. exf8=Q+ Qxf8 38. Qxg6+.
34...hxg4 35. Rxf7 Kg8 36. Ref1 and wins, as there are too many threats to meet, mainly 37. Rf8+ and 37. Qg5+.
|Mar-14-14|| ||M.Hassan: "Difficult"
White to play 33.? and White is a pawn up.
Some strange feeling informs me that the Queen should get to h6, therefore pawn movement:
And White wins!
|Mar-14-14|| ||vajeer: I wonder if 34.Bxh5 is better
34...gxh5 then Qg5+ followed by f6 looks strong
34....gxf5 then Bxf7 followed by Rxf5
|Mar-14-14|| ||Kinghunt: <An interesting idea is to look for a surprise down the line. But how do you define a “surprise”? One possibility is to be able to find a good move that the opponent, human or machine, does not see. And one way to do this is to be able search to a deeper depth than the opponent, but I don’t see how this is any different than seeing farther over the horizon than the opponent.>|
I have given some thought to this, as it may be the best way to tune engines to play well against humans. (Not that they need any help, but they may get better results by not jumping at so many incredibly hard to find threats.)
I think a "surprise" may be considered as a move that causes a drastic evaluation shift above a certain depth. Suppose an engine searches to d=24, and has two moves that it evaluates roughly the same. Normally it just picks whichever one is a centipawn higher. But I think it can do better than that. It should play the move that scores <worse> at lower depths (say, d=18).
The idea is simple - if the evaluation has risen with depth, it means the engine refuted several defensive tries - defensive tries that seemed very plausible even at d=18. Anything engine approved at d=18 is likely to tempt a GM, or at least make them spend time thinking about it.
Let's make this a little more concrete. Suppose we have the following position, black to move:
click for larger view
Let's say the two candidates are 17...d5 and 17...O-O. At d=18, we get the following lines and scores:
17...d5 18. g6 hxg6 +1.25
17...O-O 18. g6 d5 19. +1.77
Then say the engine runs to d=24, and gives the following:
17...d5 18. g6 hxg6 +1.13
17...O-O 18. Rdg1 Rfc8 +1.15
I would argue that the engine should choose the second line. Why? Because the opponent is going to be tempted by 18. g6, as was the engine at d=18. However, at d=24, it found that after 17...O-O 18. g6, it has the resource 18...Bf6!, which drops the eval to +0.78.
That is how an engine can detect "surprise" and use it to "trick" its opponents. It risks very little, as at full depth, the moves evaluate similarly anyway.
It is, of course, a matter of parameter tuning to determine when two moves evaluate "close enough" to employ this strategy without undue risk, as well as the lower search depth being compared with, and some kind of function to integrate the two.
I would appreciate any and all feedback on this idea. I may cobble together my own version of Stockfish with the move selection changed in such a manner to be deliberately difficult for humans to play against.
|Mar-15-14|| ||Patriot: <Kinghunt> That's an interesting idea. So if it happens to be losing it goes with a move that is actually better for its opponent while giving them a tempting way to screw up, since their limited search depth only allowed them to see how "great" 18.g6 is. Very clever!|
That's sort of the way strong human players play. When they are losing they may go with a crazy move that is worse but gives them the best chance to get back in the game. After all, they are only going from "lost" to "dead lost" if it doesn't work.
|May-21-18|| ||Saniyat24: agadmator made a nice video in youtube on this game...!|
|May-21-18|| ||Saniyat24: El Gatto got the big mouse...!|
< Earlier Kibitzing · PAGE 9 OF 9 ·