This is probably because of the default contempt factor. An engine's contempt factor is how many centipawns it thinks the pieces are better than the opponent's. SF 10's default contempt value is 24 while SF 9's is 20.
With the slightly higher contempt factor SF 10 tries to avoid repetitions and continue the game compared to SF9, but the higher contempt value results in excessive optimism and sometimes makes it go for risky lines. At worst, the contempt value gives an excessively high evaluation of the position compared to other engines, something SF 9 was already prone to do anyway. This can be misleading and can be harmful in some situations.
In any case, SF 10's evaluation are just too optimistic as far as I am concerned. This can be dangerous in correspondence play on sites where engine use is allowed.
A while back I entered a rapid play section (time limit 10 basic plus 1 day per move, no vacations) on Lechenicher SchachServer where they allow engine use because I wanted to test the SugaR Pro engine.
I got inferior positions so switched to Stockfish 9 and so far salvaged three games as draws, but one was beyond hope. When Stockfish 10 came out, I switched to it, but it suggested that all three of those games in which I salvaged draws were beyond hope. If I had believed its evaluations, I would have resigned!
For example, in the position after black's 8th move in the Forgacs – Maroczy game below, after 20 minutes all engines selected the correct move, but their evaluations differed considerably. Clearly, white stands better, but by how much? Actually, in the position white probably does have enough of an advantage to win, but the evaluations illustrate the point. So, for now I am sticking with Stockfish 9.
No comments:
Post a Comment