|
< Earlier Kibitzing · PAGE 1 OF 2 ·
Later Kibitzing> |
Jan-05-19
 | | keypusher: Courtesy of <Ron> here's a Daniel King video showing how SF beat A0 in just 22 moves. https://eswomen.com/id/alphazero-cr... |
|
Jan-05-19
 | | Ron: Thank you Keypusher. This game has influenced me. If I am faced with a position after black's 5th move, I intend to play h4. |
|
Jan-08-19
 | | offramp: Stockfish was a bit of a punching bag during this long match, but when he was put into SLEEP mode after this game he must have been very happy.
I lost count of the number of piece offers he made.
People who play the French Defence may have to reappraise their Umwelt: the Chatard-Alekhine looks too strong! |
|
Jan-08-19
 | | AylerKupp: The best moves in the game, by either side, were played by <book>. |
|
Jan-10-19
| | Ron: <AylerKupp: The best moves in the game, by either side, were played by <book>.> I disagree. In the game, 6. ... Bxg5 was a book move. But it appears not to be best. My Stockfish prefers 6. .. h6 for Black here.
I've played some trial games with Stockfish, one group where Black plays 6. ... Bxg5, the other group where Black plays 6. ... h6, and 6. ... h6 does seem better than 6.... Bxg5. I'm willing to bet that if either Stockfish or AlphaZero could play Black's sixth move, it would NOT play 6. ... Bxg5. |
|
| Jan-10-19 | | zanzibar: <Ron> funny you should mention that... AlphaZero vs Stockfish, 2018 (kibitz #5) . |
|
Jan-11-19
| | AylerKupp: <Ron> My comment about "book" was largely tongue in cheek. But you make a good point. I had assumed that 6...Bxg5 had been pretty much discredited since the original game with this variation, and played only by those who, seduced by the gain of a pawn, did not fully appreciate the initiative gained by White. So maybe Stockfish was a victim of a bad "book". You didn't indicate the scenario of your trial games (e.g. the time control), the number of games (to determine if the results were statistically significant), and, for that matter, what the win, draw, loss results of the trials were. I think that the numbers would be of interest to many. |
|
Jan-12-19
| | keypusher: <AylerKupp> <Ron> 6....h6 (and 6....Nc6) score better in the CG database than 6....Bxg5 (or anything else). But what is Opening Explorer, really, but a chronicle of human error? :-) Opening Explorer |
|
Jan-15-19
 | | FSR: Gotta love the theme of sticking one's bishop and knight into a pawn fork on successive moves, à la P H Nielsen vs T Hillarp Persson, 1998 and Aronian vs Anand, 2013. |
|
Apr-14-19
 | | PawnSac: < AylerKupp: You didn't indicate the .. results of the trials were. I think that the numbers would be of interest to many. > without a doubt! |
|
Apr-14-19
| | PawnSac: < offramp: Stockfish was a bit of a punching bag during this long match, > lol, yea a pretty good thrashing.
AlphaZero
+63 -10 =147
..if i counted correctly. |
|
Apr-14-19
| | PawnSac: 1. e4 thru 8...Qe7 "book"
9. Qg4 g6
10. Ng5 h6
< 9.Qg4 > is also a "book" move, which first appears
in the CG database in games of Bogo, Reti, and Euwe (1919-1920) < 10.Ng5 > A more recent entry (2001+). The old move was Nf4, then 0-0-0 (1966-2008) 11. O-O-O
This seems to be the first real departure from known praxis.
CG database gives only Bd3 (2 games). |
|
| May-09-19 | | Howard: But exactly where did Alphazero go wrong ? Daniel King analyzes this game in a recent issue of "CHess Monthly" but he doesn't really indicate where Alphazero could have improved. |
|
| May-09-19 | | Jambow: 22 moves? Was there hard drive crash... |
|
| May-14-19 | | Howard: Could be! But where did Alphazero (to repeat) go wrong ?! |
|
May-14-19
 | | beatgiant: 12...Nb6 and the later 16...Nd7 seems like loss of two moves. But the immediate 12...Nf8 doesn't look great either. How about 12...Ndxe5 13. dxe5 Nxe5, getting three pawns for a knight and breaking out of the center blockade? |
|
May-14-19
 | | Sally Simpson: ***
Nice game by S.F. (Spassky - Fischer! I can link back onto that.) And I mean that - there are a lot of tricks, traps and pitfalls just waiting for a human footprint and not all of them are beyond human calculation to dig them out OTB) it's something I'd like to see more of - unfortunately it sacced a pawn because it was told to by following book. (see notes) I wonder, if given the same position it would throw the pawn without being told too. This AylerKupp is the type of move I want them to play more often and not being told too by theory. Get them to take a pawn/piece hit especially if 20 moves down the line it's eval is just a little lower than it's first choice and go for it. 'Go for it' as in not play it against another box but show it to the human user as a possible path and I'm not just talking about opening moves. Then we get, as here. Treats Galore!
The unanswered question appears to be, 'Where did Alphazero go wrong?' You will have to ask it. (OOPS, forgot you can't) Looks like too much human interference, following book input by a human from human games. C.G. might not have any human v human games with 11.0-0-0 but I have two. Both White wins. (not as classy as this game) Maybe the Alekhine-Chatard just cannot be held.
(which I doubt but something in there sent Alpha into a nosedive.) We use to blame the 'horizon effect' but apparently that no longer exists in these top machines. Selecting plausible paths (pruning?) is the new buzz word. It got dropped into the French jungle by a human, climbed the 'wrong tree' to see where it was and went off-track. After Rd3 - (what a charming 19th century looking move that one is. It reminds me of a McDonnell vs La Bourdonnais game.) the board is buzzing with tricks. When it got to here: 
click for larger view15...axb5 you get 3 bits for a Queen and fight on and on. That way you do not become a miniature and invite attention to it. (these two beasts have played each other 200+ times. This is the only one I've played through.) Here I'm reminded of Fischer-Spassky game 11. (told you I could link back to them two) Many writers, Fine, Purdy, Reshevsky, Golombek, Alexander... are asking why Fischer did not resign earlier, Fischer played on for a few extra moves so it would not be a 25 moves miniature and also not the shortest loss of the match - that was game 5 (a 27 move Fischer win). Something else computers need programmed into them, an ego. *** |
|
| Oct-27-19 | | Chesgambit: 11. 0-0-0 equal position ( very high depth analysis |
|
| Apr-08-20 | | Everett: 3..Nf6 is the mistake! |
|
| Apr-08-20 | | Everett: <Here I'm reminded of Fischer-Spassky game 11. (told you I could link back to them two) Many writers, Fine, Purdy, Reshevsky, Golombek, Alexander... are asking why Fischer did not resign earlier, Fischer played on for a few extra moves so it would not be a 25 moves miniature and also not the shortest loss of the match - that was game 5 (a 27 move Fischer win).> haha |
|
| Apr-17-20 | | ldrs: The best game I’ve ever seen |
|
Apr-17-20
| | AylerKupp: Yes it was. But I get another opportunity to once again repeat my observation. With AlphaZero's estimated (by me) 80X computational performance advantage over Stockfish my only surprise is that the thrashing wasn't even worse. Based on DeepMind's data for a 100-game match and a 1/80X time control disadvantage for AlphaZero (in order to equalize the two engine's estimated computational performance), for a 220-game match this would have had Stockfish winning by a margin of +81, -11, =128. An even bigger thrashing. When one engine executes in hardware having a substantial computational advantage over the hardware available to the other engine, the actual match results are pretty much irrelevant if you're trying to compare the relative performance of the two chess engines. |
|
| May-20-21 | | DouglasGomes: The reverse ended in a draw: AlphaZero vs Stockfish, 2018 |
|
| May-24-21 | | DouglasGomes: 4... Be7 is a positional mistake, French players should just give up on this and play 4... dxe4. I am too lazy to investigate whether the engine's 8... Qh4 is a improvement over 8... Qe7, but things don't look good after 8... Qe7, just suffering and waiting to see if your opponent wins or draws. Nowadays we know better that the advance of the rook pawn is a powerful idea, so I would guess the inclusion of h4/h6 favors White. These lines after Be7 are rarely played by elite GMs, so I would also suspect there are many theoretical improvements to be found by ambitious chess players. <beatgiant>'s 12.... Ndex5 13. dxe5 Nxe5 14. Qf4 (Qxg5 15. Qxg5 hxg5 16. Rxh8+ Kd7 17. Re1 two pawns for the exchange but horrible for Black) f6 15. Nxc7+ so much for three pawns, Black can get his piece back, but 15... Qxc7 Qxf6 looks like all Alekhine-Chatard positions, White's is down a pawn but also winning. |
|
May-25-21
 | | perfidious: <DouglasGomes: 4... Be7 is a positional mistake....> This is bust-a-gut stuff.
One takes humour where one must, though. |
|
 |
|
< Earlier Kibitzing · PAGE 1 OF 2 ·
Later Kibitzing> |
Kibitzing
sac: 20.Nce4