A tactical puzzle that Stockfish missed.

stiggling
PolarPhoenix wrote:

i think stockfish makes its moves based on win percentage, and in the initial moves, white has a low win percentage

I guess you mean Monte Carlo tree search? It doesn't use that.

PolarPhoenix
stiggling wrote:
PolarPhoenix wrote:

i think stockfish makes its moves based on win percentage, and in the initial moves, white has a low win percentage

I guess you mean Monte Carlo tree search? It doesn't use that.

ok didn't know, thanks for telling me though!

jonathanpiano13
I wouldn’t call it simple.

It’s all a matter of hardware I guess
drmrboss
jonathanpiano13 wrote:
I wouldn’t call it simple.

It’s all a matter of hardware I guess

I stand corrected, it is not simple.It is more like search algorithm issue than hardware( in theory, if you have bad hardware, longer time control will find the solution, if you have bad software, it is almost impossible to find it. For example, in previous times, you might see older engines saying +5 or +9 in dead draw endgames).

So, how this happens, according to material count, white is losing. In search of all possible lines 99/100, Stockfish will see losing moves, but in 1/100, he sees a drawish line. According to Alpha Beta pruning, programmers cut the lines that seemed losing in shallow search.

null

If you look at diagram , there are trees that an engine dont search anymore. And the answer is deep down in the line of ( for example , G in the diagram).

Cos, 1. Nc8 missed the chance to get a forced draw, and SF thinks that line is losing in shallow search( so SF discarded that line).

Unless, a programmer write another patch or improve search, it is unlikely to see.