| Jan-02-17 | | Murky: 18. f5 is a nice tactical shot, typical of Persson's style. Yet the Stockfish analysis at chessbomb dot com initially rates f5 as a mistake. But then as the players follow the exact lines given by Stockfish, 18. f5 looks better and better until white is practically winning. What does Persson know that the best chess computers can't quite figure out? Perhaps machine calculation isn't everything, and human intuition is sometimes as good or better!? Score one for the human race. |
|
| Jan-02-17 | | ChessHigherCat: A master of the Viking attack! Tiger or Person?: only his zoo-keeper knows for sure... |
|
| Jan-07-17 | | Eyal: <18. f5 is a nice tactical shot, typical of Persson's style. Yet the Stockfish analysis at chessbomb dot com initially rates f5 as a mistake. But then as the players follow the exact lines given by Stockfish, 18. f5 looks better and better until white is practically winning. What does Persson know that the best chess computers can't quite figure out? Perhaps machine calculation isn't everything, and human intuition is sometimes as good or better!?> One obvious reason is that the engine analysis provided at chessbomb (https://www.chessbomb.com/arena/201...), at least for non-premiums, is very shallow - according to the data listed there, something between 15 and 17 ply-depth. In order to get a serious evaluation of such complex positions you need 25 ply at a minimum, preferably 30+, so what chessbomb provides is very far from a display of Stockfish 8's real strength. But there's an additional factor - standard engine analysis evaluates positions based on best moves by both sides regardless of the practical difficulty of finding them, and here even with a deeper analysis Stockfish doesn't think the sac is so great, because with best defense Black can hold on. However, this doesn't take into account practical considerations of human play. For example, after 22.Rxf7 the engine wants to defend with 22...Rh7! 23.e6+ Nxe6 24.Nf3 Ra6 25.Nf4 Nc8 26.Ne5+ Ke8 and evaluates the position as equal, but this hardly looks as likely human play. At any rate, a great attacking game by Hillarp Persson. |
|
| Jan-07-17 | | izimbra: <Eyal: <18. f5 is a nice tactical shot, typical of Persson's style. Yet the Stockfish analysis at chessbomb dot com initially rates f5 as a mistake. But then as the players follow the exact lines given by Stockfish, 18. f5 looks better and better until white is practically winning. What does Persson know that the best chess computers can't quite figure out? Perhaps machine calculation isn't everything, and human intuition is sometimes as good or better!?> One obvious reason is that the engine analysis provided at chessbomb (https://www.chessbomb.com/arena/201...), at least for non-premiums, is very shallow - according to the data listed there, something between 15 and 17 ply-depth. In order to get a serious evaluation of such complex positions you need 25 ply at a minimum, preferably 30+, so what chessbomb provides is very far from a display of Stockfish 8's real strength.> That can always happen, but I see Stockfish wanting black to play <20...Kc8> and also liking a few other moves better than the game continuation. |
|
| Jan-07-17 | | Eyal: <That can always happen, but I see Stockfish wanting black to play <20...Kc8> and also liking a few other moves better than the game continuation.> I was referring mainly to the evaluation of the key move 18.f5, which is what <Murky> was talking about; but if we're talking about other moves, I also see that the engine evaluation there completely misses that 22...Nxf7 is a critical mistake (again, not such a big surprise at mere 15-ply depth). |
|
| Jan-07-17 | | izimbra: <Eyal: <That can always happen, but I see Stockfish wanting black to play <20...Kc8> and also liking a few other moves better than the game continuation.> I was referring mainly to the evaluation of the key move 18.f5, which is what <Murky> was talking about; but if we're talking about other moves, I also see that the engine evaluation there completely misses that 22...Nxf7 is a critical mistake (again, not such a big surprise at mere 15-ply depth).> I thought you were saying that Stockfish missed what a great move <18.f5> was, even though playing out Stockfish preferred lines leads to a big White advantage. I may have misunderstood the point, but that is the idea I investigated, and it seems to me that Stockfish doesn't think <18.f5> is that great because it has Black equalizing in those lines. Stockfish likes <18.Qf2>. It seems to be only later poor choices by Black that led to the big White advantage in the continuation. |
|
| Jan-07-17 | | Eyal: <izimbra> I suppose I could have expressed myself more clearly... 18.f5 gets a better evaluation with a deeper analysis than the one given at chessbomb, but I also wrote in my first post "even with a deeper analysis Stockfish doesn't think the sac is so great, because with best defense Black can hold on". Indeed, because 18.f5 leads to equality with best play by Black Stockfish prefers other moves that are evaluated as leading to a small advantage (but not nearly as forcing). Though again, from a human-practical viewpoint Hillarp Persson's choice may have been much more challenging. |
|
Jan-07-17
 | | tamar: Stockfish can often defend indefensible positions if given enough time and cores.
It held the Black side of a Fried Liver attack versus Houdini (game 84)http://tcec.chessdom.com/archive.php
where it was saddled with the position 1 e4 e5 2 Nf3 Nc6 3 Bc4 Nf6 4 Ng5 d5 5 exd5 Nxd5?! 6 Bxf7+ Kxf7 and its first choice after move 11 Qe2
click for larger viewIn this instance, Stockfish had 44 cores and three hours of time, so it had a mind-boggling depth. |
|
| Jan-07-17 | | izimbra: Black sets up to castle KS, but delays actual castling to generate QS play. White reacts with KS pressure, making castling to risky. Black panics with the ugly looking <15..Kd7> that almost invites problems, though the engine thinks it can hold. Stockfish is a lot better at chess than me, so I can learn from looking at how it would proceed and defend instead. It opts for <15...b4 16.Bd1 b3 17.Qg3 g6 18.Be2 Bf8 19.Be3 Na4 20.Rfb1 Qd7 21.Nf4 Ne7 22.Qh3 Nf5> which is even and may lead to a 3-fold repetition. There are 2 combos of ideas in that - tactics involving b4 and Na4 on the QS and defense with g6, Bf8, and Ne7 on the KS. |
|
| Jan-07-17 | | Eyal: <tamar> In this context, here's a quote from a recent interview with Svidler: <The main change [produced by computers over the game], a completely fundamental one, is the new evaluation of defensive resources. Previously a huge amount of analysis stopped in the region of move 17-19 with the symbol for “with an attack” – that was really good. Now, having worked with computers for decades, you understand very well that “with an attack” isn’t an evaluation. The computer defends positions that previously, having taken a single glance at the board, you’d consider to be mate. You didn’t know how, but there was no need – you’d find it at the board. But the computer finds such resources that people as well have started to evaluate defensive resources completely differently. It’s become clear how balanced the game is and how difficult it is to upset that balance.> (https://chess24.com/en/read/news/sv...) |
|
Jan-07-17
| | tamar: Nice quote from Svidler.
Stockfish and Komodo, and to a lesser extent Houdini, have a propensity to shut down any opening once the danger signals of attack are triggered. Finding a white advantage is becoming a problem! Steinitz recognized that "many tempting and successful sacrifices turn out to be incorrect. I came to the conclusion that sound defense demands far less expenditure of energy than attack." These days showing an engine a move like 18 f5 is apt to bring a swift 0.00. It is 99% winning, but the engine finds the 1%. |
|
|
|
|
Kibitzing
sac: 19.Bxf5
