Because... because... that would spoil all the fun :P

<That's a stupid reason not to look at a great game of chess.>

Maybe you should try Fool's Mate sometimes, <keycrusher>. It's also easy to remember..

Besides, you are cherry licking. This whole project is a scam and you know it (that I'm sexy)>

Brain-dead? Well, actually we are speaking of a dead brain. Pretty close but no marmalade.

If you weren't that lame, you had checked this variation at cg here and noticed that there are 2,746 games (1897-2017) after 6. Nf3 and only 15 games with 6... cxd4, 0.05% (although I think some more if one counts the transpositions).

2,510 games with 6... Nc6

But that's not all; 6... cxd4 has 46.7% losses for Black. Which is the high score in this variation (6. Nf3).

In 120 years 15 games of theory. You probably think that's romantic. And what do you know?

They all played 7. Nxd4

Ha ha ha isn't that ironic? And yes, the experts always think they know everything obvious... until it's obvious that they know very little about any subject en detail.

Speaking of details:

http://www.computerchess.org.uk/ccr...

Now, what do you know? There are 9337/C11 games from 2006 until now. Hundreds start with 6. Nf3 but none of dem engees play 6... cxd4, check it for yourself. Have fun! NONE OF THEM!!!

I *wonder* why...

But... eh.. Stockfish plays 6... cxd4 and you are going to tell us how wonderful that is and oh, what a great game! And call ME an idiot (not that it's of any importancy; I probably am).

I hope you DO understand that these are nice puzzles, show how most extreme inactivity doesn't work in chess, you don't say, but the games DO NOT HAVE any relevance to any chess theory whatsoever. Who knows what the future may bring. But that's another question. So far.

Consider this recent game from the European Team Championship: Y Kuzubov vs V Sanal, 2017 - or more specifically this position:

click for larger view

Even at a depth of 60 ply, droidfish v 1.71 (a packaging of Stockfish for cell phones, using a build of Stockfish from september 2017) wants to play the blunder Rxf1+ followed by Bxa3 - which is a dead draw, despite the position before Rxf1+ being completely won for black (as demonstrated by Sanal in the game). Why? Horizon effect which causes the engine to choose the *local* maximum within its horizon, in other words the blunder Rxf1+ which wins the most material.

To quote the AlphaZero paper:

<AlphaZero evaluates positions using non-linear function approximation based on a deep neural network, rather than the linear function approximation used in typical chess programs. This provides a much more powerful representation, but may also introduce spurious approximation errors. MCTS averages over these approximation errors, which therefore tend to cancel out when evaluating a large subtree. In contrast, alpha-beta search computes an explicit minimax, which propagates the biggest approximation errors to the root of the subtree.>

Hence, for certain kinds of positions, an MCTS will provide better results, as it doesn't prune away the winning move/line the way a "standard" mini-max search may happen to do - or, like here, propagate the huge approximation error -4 point something (after Rxf1+) to the root of the search tree. (The correct evaluation of Rxf1+ is 0.00.)

This is basically the same kind of issue that makes it hard for (older versions of) Stockfish (and friends) to find/see the Bxg6 sacrifice in the current game - even though human, strong players at IM level (or even lower) will understand that Bxg6! is a very good move. To avoid blunders like Rxf1+ in the above position, the mini-max algorithm needs additional tinkering.

Cherry licking? Really?

<keypusher....You post a lot of crap on many topics....>

[Fritz 10]: 1. d4 e6 2. e4 d5 3. Nc3 Nf6 4. e5 Nfd7 5. f4 c5 6. Nf3 cxd4 [last book move] 7. Nb5 Bb4+ 8. Bd2 Bc5 9. b4 Be7 10. Nbxd4 Nc6 11. c3 a5 [11 ... Nxd4 12. Nxd4 a5 13. Bd3 ⩲] 12. b5 Nxd4 13. cxd4 Nb6 14. a4 Nc4 15. Bd3 [15. Bxc4!? dxc4 16. 0-0 ±] Nxd2 [=] 16. Kxd2 Bd7 17. Ke3 b6 [17 ... 0-0 18. Kf2 =] 18. g4 h5 [18 ... 0-0 19. g5 ⩲] 19. Qg1 [19. g5 0-0 ⩲] hxg4 20. Qxg4 Bf8 21. h4 [21. Qg5 Qc7 =] Qe7 22. Rhc1 [Traps the king in the center] g6 [22 ... Rc8 23. Rxc8+ Bxc8 24. Rc1 =] 23. Rc2 Kd8 [Black loses the right to castle - 23 ... Ra7 24. Rac1 ⩲] 24. Rac1 Qe8 25. Rc7 Rc8 26. Rxc8+ Bxc8 27. Rc6 Bb7 28. Rc2 Kd7 29. Ng5 Be7 30. Bxg6 Bxg5 31. Qxg5 fxg6 32. f5 Rg8 33. Qh6 Qf7 34. f6 Kd8 35. Kd2 Kd7 36. Rc1 Kd8 37. Qe3 Qf8 38. Qc3 Qb4 39. Qxb4 axb4 40. Rg1 b3 [40 ... Bc8!? 41. Kc2 Ke8 ⩲] 41. Kc3 [ ±] Bc8 42. Kxb3 Bd7 43. Kb4 Be8 [43 ... Ke8!? 44. Rc1 Rh8 ±] 44. Ra1 [White prepares the advance a5] [+-] Kc7 45. a5 Bd7?? [Black crumbles in face of a dire situation - 45 ... bxa5 46. Rxa5 Kb7 +-] 46. axb6+ [+-] Kxb6 47. Ra6+ Kb7 48. Kc5 Rd8 [48 ... Rh8 does not help much 49. Ra2 Rxh4 50. b6 +-] 49. Ra2 Rc8+ [49 ... Be8 is no salvation 50. b6 Rc8+ 51. Kd6 Kxb6 52. Rb2+ Ka7 53. Ke7 +-] 50. Kd6 Be8 51. Ke7 g5 [51 ... Kb6 cannot undo what has already been done 52. Kxe6 Rc7 53. Ra8 Bxb5 54. Rb8+ Ka5 55. f7 Rxf7 56. Kxf7 +-] 52. hxg5 1-0.

I understand the irony of using an engine to review a computer game. ;>D

The AlphaZero - Stockfish (2017) computer match was a convincing lopsided victory for the newcomer with 28 wins, 72 draws and no losses; the database has ten of AlphaZero's wins. The French often gives White a clear space advantage out of the opening, but this game demonstrates a gradual process of pushing Black off the board, who is helplessly reduced to shuffling pieces at the end. People noticed 17. Ke3 as a quaint computer move but note another quaint computer move after 15 ... Nxd2:

click for larger view

16. Qxd2?? is out due to 16 ... Bb4, but I'll bet 100 out of 100 humans would automatically choose 16. Nxd2 followed by 0-0, making AlphaZero's 16. Kxd2!? and 17. Ke3 quite inspired. Other computer moves are 19. Qg1 along with the Pawn throwaways 40 ... b3 & 51 ... g5. Notice Fritz the anti-human cyborg awards no "!" to 30. Bxg6 whereas many commentators would say "!!" without question. This endgame is not exactly an easy win but looks like it when the White King marches in on the dark squares.

We all know that computers can beat us at our own game. Without delving into Terminator philosophy, this match result raises the real possibility that computers will "solve" chess within our lifetimes, or at least make computer matchups evenly played draws by definition. Disturbingly, it also increases the specter of cheating with computer help ruining OTB play.

Chess fans around the world are now anxiously awaiting AlphaZero's book on how to defeat the French Defense.

For trivia buffs, this is the first computer game to earn a spot in the Best Games contest. Since you asked, see for yourself:

Game Collection: 2004's Greatest Chess Games

Game Collection: 2005's Greatest Chess Games

Game Collection: 2006's Greatest Chess Games

Game Collection: 2007's Greatest Chess Games

Game Collection: 2008's Greatest Chess Games

Game Collection: 2009's Greatest Chess Games

Game Collection: Best Games of 2010

Game Collection: Best Games of 2011

Game Collection: Best Games of 2012

Game Collection: Best Games of 2013

Game Collection: Best Games of 2014

Game Collection: Best Games of 2015

Game Collection: Best Games of 2016

click for larger view

White to move

1) +0.80 (38 ply) 29.Qg5 Rh7 30.Qf6 Kd7 31.Ng5 Rxh4 32.Nxf7 Rh3+ 33.Ke2 Rh4 34.Kf1 Qxf6 35.exf6 Rxf4+ 36.Rf2 Rxd4 37.Ne5+ Kc7 38.Nxg6 Bb4 39.Bc2 Rd2 40.Rxd2 Bxd2 41.f7 Bh6 42.f8=Q Bxf8 43.Nxf8 e5 44.Ke2 Kd6 45.Ng6 d4 46.Kd3 Bd5 47.Nf8 Bf7 48.Kd2 Bd5 49.Nh7 Bf7 50.Ng5 Bc4 51.Bd3 Bb3

2) +0.25 (37 ply) 29.Ng5 Bh6 30.Kf2 Kd7 31.Qg3 Rc8 32.Kg2 Rxc2+ 33.Bxc2 Kc8 34.Bd1 Kb8 35.Qc3 Bf8 36.Kf2 Bg7 37.Kg3 Bh6 38.Be2 Bf8 39.Bf3 Bh6 40.Qc1 Bg7 41.Bg4 Bf8 42.Qc3 Ka7 43.Bd1 Bg7 44.Bf3 Kb8 45.Qc2 Ka7 46.h5 gxh5 47.Qh7 h4+ 48.Qxh4 Qb4

3) +0.25 (37 ply) 29.Qg3 Kd7 30.Ng5 Bh6 31.Kf2 Rc8 transposes to line 2).

1. I haven't seen anyone here consider 17. Ke3 as a "quaint computer move". I for one expressed fascination with the Kxd2 and Ke3. If played by a human, I would have called it genius.

the peer reviewed Science paper reports that a total of 1000 games were played against the Stockfish version that won the season 9 TCEC world championship. the final score for AZ was: +155:-6:=839.

in these games:

1. AZ puts a high value on space for its pieces.

2. a major tactical theme is restricting or enclosing opponent pieces, rather than attempting to exchange off or capture them.

3. a major performance theme is that AZ does significantly less well when playing black, in any position; the greatest discrepancies between win ratios on opposite sides of the board occur in the caro-kann, the french, and the spanish (e.g., these were the openings that it most consistently demolished as white and most consistently lost as black). this implies that it preferred openings where the benefit of tempo is most consistently preserved by white.

4. the AZ logic is fundamentally positional rather than tactical (it does not go in for the "brilliant sacrifice"). AZ moves tend to leave the opponent without good replies, rather than force exchanges to modify the position. (this is consistent with #2).

5. AZ tends to win with fewer pieces or less material on the board, and it seems especially casual about sacrificing pawns for lines of attack; otherwise, grabbing material is not really attractive to it, compared to freedom of movement for the pieces that are left. it's not what it has but how it can be used (again, consistent with #2).

in all, the AZ games I have seen most resemble the chess logic found in Morphy.

the development strategy for AZ, letting the neural net learn by "playing against itself" and playing without an opening book is intriguing, because AZ seems not to have done well when presented with an established opening position as the start of a game. this is a standard challenge in computer v. computer evaluations, and was done again here for comparative purposes. these positions tended to produce many draws, and as a group tended to be less and less preferred by AZ as its "training" progressed. this suggests:

6. the chess logic in AZ is more "organic" and a continual elaboration of opening objectives which the program cannot retrospectively infer from a position given.

in general AI terms, #6 implies that it is difficult to infer a strategic idea from a static chess position; this may partly explain the "amazement" at some of the AZ "brilliancies". so finally:

7. a key factor seems to be that AZ did not perceive threats where there were none.

#7 is often described as "risky" play by commentators who are clever at finding threats. finally: