oh yeah @ubdip that's a problem I have noticed in a lot of engines, they don't want to end the game
That's just in the PV.
The moves shown several lines into a PV are not particularly reliable. It's not uncommon for them to be outright blunders; of course, here it's only "weird".
Basically, you should take moves further into a PV with several large grains of salt.
The deeper the search, the more you can trust, but there's not a hard rule about this.
If you actually go to the position where the last pawn could be taken, SF wants to take it starting from depth 2 and maintains it all the way after that.
In general, though, there will be some oddities that crop up (from a human perspective) when there are ties for the best move.
An engine may always pick one that seems strange to us.
The moves shown several lines into a PV are not particularly reliable. It's not uncommon for them to be outright blunders; of course, here it's only "weird".
Basically, you should take moves further into a PV with several large grains of salt.
The deeper the search, the more you can trust, but there's not a hard rule about this.
If you actually go to the position where the last pawn could be taken, SF wants to take it starting from depth 2 and maintains it all the way after that.
In general, though, there will be some oddities that crop up (from a human perspective) when there are ties for the best move.
An engine may always pick one that seems strange to us.
As previously mentioned, engines (especially if they have the "contempt" score turned "on," and/or are trying to flag their opponent) sometimes don't end a game a human player would. Take this for instance:
lichess.org/analysis/7k/8/8/8/8/1K6/1p6/8_w_-_-
With white to move, any human player would capture the pawn and get on with his life. Computers, on the other hand, don't care about the Giants game or getting to bed on time. I have been running computer vs. computer matches and seen many times, moves similar to Ka2, or Kc2. Tablebase play might be different, but without it, the computer doesn't care if Black queens, so long as it can force the draw thereafter.
lichess.org/analysis/7k/8/8/8/8/1K6/1p6/8_w_-_-
With white to move, any human player would capture the pawn and get on with his life. Computers, on the other hand, don't care about the Giants game or getting to bed on time. I have been running computer vs. computer matches and seen many times, moves similar to Ka2, or Kc2. Tablebase play might be different, but without it, the computer doesn't care if Black queens, so long as it can force the draw thereafter.
I'll take a look at Stockfish's PV-extension (use longest PV) code since for won/drawn positions the shortest PV may be useful.
This topic has been archived and can no longer be replied to.