And SF doesn't know bishop+knight checkmate.
Stockfish is Dumb
Sort:
Feb 6, 2018
0
#3
USArmyParatrooper wrote:
It assesses White is winning here. The position is 0.00
It's not just Stockfish. You'd probably be hard pressed to find any engine that realizes it's a draw.
Yet a 1400 player can see in 3 seconds it is a draw. What if earlier SF had a choice of trades and could have had a bishop or knight and chose the bishop because it is an open position and the bishop is worth a few centipawns more?
Feb 7, 2018
0
#5
Chesserroo2 wrote:
Yet a 1400 player can see in 3 seconds it is a draw. What if earlier SF had a choice of trades and could have had a bishop or knight and chose the bishop because it is an open position and the bishop is worth a few centipawns more?
In full disclosure, I did not come up with this on my own. Simon “Ginger GM“ has an interesting video on Chess engine quirks. For the heck of it I decided to set up the position and see what the computer thought of the position for myself, and voilà he was right.
But truth be told, if I managed to force Stockfish into such a position I would feel like a God.
Feb 7, 2018
0
#6
Not a common situation, but Stockfish reckons it is doing great if it has 2 or more bishops on the same colour squares! Believe me, Mr. Stockfish, if you have 9 of those you are still not going to win!
Elroch wrote:
What if they trade themselves for the center pawns at just the right time so their king has opposition and can come in? Push the center pawn, then nab the g pawn. Would the king lose opposition?
Not a common position, but Stockfish reckons it is doing great here. A slightly more serious one is that even with an hour of thinking time on multiple cores it can't work out that if it has two bishops of the same colour, it isn't winning.
Elroch wrote:
Better yet, with two of the same color, target the lead pawn from both sides so the king has to move away from it, then double hit it, trading one bishop for two pawns.
Not a common position, but Stockfish reckons it is doing great here. A slightly more serious one is that even with an hour of thinking time on multiple cores it can't work out that if it has two bishops of the same colour, it isn't winning.
Elroch wrote:
Put one bishop on g2 and get the other to g7, then take on e5 when the king steps back. Two same color bishops is a win for White.
Not a common position, but Stockfish reckons it is doing great here. A slightly more serious one is that even with an hour of thinking time on multiple cores it can't work out that if it has two bishops of the same colour, it isn't winning.
So you found a type of position famous only because engines can't solve it?
If it was dumb then this wouldn't be news.
If it was dumb then this wouldn't be news.
Feb 7, 2018
0
#11
Engines simply aren't too good at calculating fortresses. For a human, however, it's easy to see no breakthrough for white; even trying w/ his bishop doesn't work as black can simply repeat w/ Kc7 and Kd6 to score a draw.
If you open a respectable number of threads, and see all of them stuck at precisely the same evaluation (+2.62 in your case) then you may well start thinking about a fortress being present.
Feb 7, 2018
0
#13
Chesserroo2 wrote:
Elroch wrote:
Not a common position, but Stockfish reckons it is doing great here. A slightly more serious one is that even with an hour of thinking time on multiple cores it can't work out that if it has two bishops of the same colour, it isn't winning.
Put one bishop on g2 and get the other to g7, then take on e5 when the king steps back. Two same color bishops is a win for White.
Two (or generally more) same coloured bishops is usually NOT a win without co-operation.
Eg try 3k4/8/2B5/3B4/4B3/8/8/4K3 w - - 0 1 in a tablebase.
(I am unsure why people have copied the starting board from a draft of my post: this was only there because of what seems like a remaining bug in the diagram editor).
Feb 7, 2018
0
#14
pfren wrote:
If you open a respectable number of threads, and see all of them stuck at precisely the same evaluation (+2.62 in your case) then you may well start thinking about a fortress being present.
Someone should tell the engines that!
Elroch έγραψε:
pfren wrote:
If you open a respectable number of threads, and see all of them stuck at precisely the same evaluation (+2.62 in your case) then you may well start thinking about a fortress being present.
Someone should tell the engines that!
Sometimes it happens in other cases- e.g. when the engine thinks that the actual winning move degrades the position, and instead of picking it, it aimlessly moves the pieces back and forth to maintain the "advantage"...
Here is a correspondence engines-on game I played recently. First, I avoided established theory by playing a "second rate" opening, and after achieving a respectable attacking position (14.Qh5) I had to ignore the engine suggestions, which were misleading, and form a (slow, rather boring, but extremely effective) plan by my own, using the engine just for checking all kinds of tactical tricks. It worked very well- the bishop pair totally dominated the board.
In this position stockfish evaluation says king c7 is best which loses a bishop
It assesses White is winning here. The position is 0.00