I was doing KQKN practice
lichess.org/practice/checkmates/piece-checkmates-ii/Rg2cMBZ6/duxBgZ8tI reached this position:
And the computer moved Kc5 (as shown to the right) allowing me to simply take the knight. No human would allow that, when you can interpose instead and the checkmate sequence following that doesn't seem trivial at all to me. A human would defend "better" in the sense that the theoretical length of the checkmate sequence doesn't really matter past a few moves, but the clarity of the plan does.
On top of that, clicking on "analysis board" brings me to the original problem position rather than the one I'm at.
I've done the problem a couple times but I'm not satisfied that a human couldn't be much trickier OTB than stockfish is here. I see this must be hard to simulate but it would be nice if you could at least make the computer try alternative lines.
Regards,
b.a.i.l.
I'm just a patzer, but I'm confused: did you have some move other than Kc5 that would *not* have allowed the queen to take the knight?
@fncll Yes. It seems that the move Nb5 not only gets out of check, but also saves the knight.
Kc5 because after Nb5 Qb4 wins even faster.
yeah it wins faster *in theory* it's a mate in 6 which isn't obvious at all - allowing simplification to a KQK ending is silly
Stockfish doesn't care about what a human would do, or what "makes sense." It gives each move a score and the best score is its choice. That it.
A human might think "OK, I'm lost, so I have to find the trickiest moves even if they're not the best moves." Stockfish does nothing like that. It's a notion that simply doesn't exist in its world.
It's entire evaluation function is based on the assumption that the opponent will play the best move.
I suppose they could add a "play for the cheapo" switch, and when the opp's second-best move is sufficiently worse than the best one then Stockfish chooses that line instead of the objectively best move.
>Stockfish doesn't care about what a human would do, or what "makes sense." It gives each move a score and the best score is its choice. That it.
I know, I think I clarified that sufficiently in the message. This is probably hard to do - engine modifications or some rule system on top - so one thing that could be done is allowing for alternative lines on request somehow.
I mean, I know it's probably not high in the priorities - but as is, it's quite useless for actual practice of this endgame because no human will simplify the endgame for you voluntarily for abstract theoretical mating length considerations. The mate sequence after the interposition is quite counter-intuitive and I'm not sure I can do it under time pressure OTB.
PS: just want to clarify that I think the site is great, I only registered recently and I love many of the features. This is just one thing that could be improved - I have many other suggestions btw. I don't want to sound demanding though.
This is funny, because it is xactly the difference between 'distance to mate' and 'distance to conversion' play. Old end-game tables were using DTM, but this sometimes erred with the 50-move rule. So most engines have switched to using DTC EGTs (or actually DTZ50, which is oly different when there are Pawns around). And most of the time this leads to people complaining about exactly the opposite, that the computer makes stupid conversions, delaying the mate.
Indeed it's ironic! "Thankfully" Lichess AI/analysis doesn't use endgame tablebases so endgame puzzles (if any) would play "optimal" moves.
