Well , I remember my first chess computer , a NOVAG JADE in the 1980s. With an incredible (then) 32 KB of RAM and 2,000 half moves opening book. It could even solve mate in 2 problems!
Its rating was estimated at 1900 uscf , but it was lower...
Those were the days....
You see , Kasparov hadn't been defeated yet , and there was still hope for the human race!!!!
@mdinnerspace do you have the solution for the puzzle?
yes :-) I'm "smarter" than SF ! LOL
It starts with 1. f3 a4 2. Kf2!
white plays f3 Kf2 Kg3 Kxh3 Kxg2 with a fortress. It's a silly position. But stockfish does find it, it just doesn't evaluate it as a draw.
Is it really so silly? Several fortress positions were just played, behind major material vs. a Queen by GM's in the last events to make a draw. SF finds it alright, but won't play it as it has no "positional" understanding, only a material advantage evaluation.
Fortress recognition is theroretically the biggest problem in computer chess, because they don't get it. Practically it's no problem at all because no game ends in a fortress virtually.
So f...ing what? Elo 3300, 3400, 3500? We are all clowns as Nakamura pointed it out.
That engines have no positional understanding isn't really true anymore. Sure it doesn't evaluate positions the same way humans do but they do have positional awareness. F.ex. piece values vary depending on its position on the board.
CM Sarg0n edited3 minutes ago #36
Fortress recognition is theroretically the biggest problem in computer chess, because they don't get it. Practically it's no problem at all because no game ends in a fortress virtually.
I guess the very recent match between Ivanchuck and Wei Yi, where Ivanchuck was up a Queen and pawns vs. Rook and Knight and pawn, with SF screaming Ivanchuck was winning with a big + score and Wei Yi's beautifully created fortress... never happened.
@lovlas If you missed post #23, it is an excellent summary of computer evaluation. In assigning a value to the pieces their position on the board most definitely is considered. How many squares are being controlled, which squares, it's mobility, threats etc. all give a +/- to it's value. But is the comp "understanding" a position in the human sense, or is it simply crunching numbers the programmer assigned it to search for?
#39 this is branching into a philosophical question. It has nothing to do with wether engines are able to see positional nuances, but if they are at all capable of understanding anything. If you want to go even further you could ask the same about humans. Are we able to understand positions, or is it just calculation mixed with experience? Imo these questions are irrelevant and we should stay on topic.
