I saw a video recently that said Komodo's position evaluations are more reliable than Stockfish, although that video was about 3 years ago so the latest Stockfish might be better.
Engines, KID, French, etc.
Sort:
Sep 13, 2021
0
#3
I mentioned this in a reddit discussion... that you can't blindly trust the engine for openings, and you have to do things like actually play the engine's main line out to see whether it changes (not the only way, but a basic way to check, that sometimes takes a long time).
A bunch of noobs downvoted me... so I don't go there anymore.
Mediocrity rules when you let average people vote.
IM John Watson, in a Dec 2020 Chess Life review of GM Giri's Chessable course on the French Defense writes:
I’ve spent a great deal of time reviewing key parts of his repertoire and subjecting it to engine analysis (primarily Stockfish and Leela). The core of the repertoire is the Winawer Variation, or 3. Nc3 Bb4. Engines are notoriously prejudiced for the White side of the main Winawer variations because they value the combination of more space and the bishop pair highly. After many hours of feeding the engines the lines from Giri’s repertoire, I consistently found that positions the engine initially assessed as +1.0 (or sometimes even higher) turn out perfectly well for Black with best play by both sides. (This engine bias is true of a fair number of openings, by the way, but especially so for closed and semi-closed ones like the French).
IM Sam Collins, in his book, The KID: Move by Move, 2017, writes: "This touches on another unusual aspect of the King’s Indian, namely that computers are of limited assistance (they tend to assess most positions in the Classical Variation as winning for White because Black’s standard attack takes so many moves to execute that the mate does not appear on the engines’ horizons)."
Collins, along his descriptions of the games, mentions the engines a few times, sometimes agreeing, sometimes disagreeing. In one position however, he notes: "For a surprisingly long time my engine thinks White is fine here, which shows just how poorly they cope with King’s Indian positions (compared to their prowess in every other type of position). To most experienced King’s Indian players it’s immediately obvious that White can’t cover the key squares around his king, no matter how many extra pieces he collects."
I agreed with Collins -- black looks like he's crushing (see p86). Truth be told, I took the position and put it into chess.com's engine (Stockfish 12) and it gave black a 4.5 pawn lead. Have engines improved that much since he wrote the book? (Note: the position is "r4rk1/2N3bp/1p1p1n2/3Pp3/PBP1Pp1n/5Ppq/2Q5/R2BNRK1 b - - 0 1". I didn't think it was worth including here, because the position is after 23 moves of KID, while I'm focusing more on the opening 10-15 moves or so).
Hanging Pawns, in his intro video to his KID series, mentioned that many times the engines shows black at -1.0 when it was a good position for black, and added that engines are notoriously bad for KID and Pirc.
@darkunorthodox88 (NM Miguel Fonseca?) notes, in an chess.com thread discussing KID theory at odds with engines, noted:
Engines tend to value "guaranteed" space over potential space. in defenses like the KID and in some variations of the french, getting "the second wave" of the territory expansion gives you an open hand in what to do with said space (for example, get a favorable rook file or knight outpost or just strangle your opponent on said side, to make a passer in the third rank etc).
Its also why engines engines dont understand defenses that are deemed fine like the czech benoni. While the potential for a big squeeze is less, the flexibility of the counterpunches is underestimated. the engine has to brute force search ALOT to reduce such openings to a fair evaluations. Humans from Experience and intuition know there is more poison to them, then it first appears.
iChess, as an opening line to one of their courses writes: " "Kasparov and Fischer loved it. Engines hate it. The King's Indian Defense is an enigma"
A 2019 Article, Fat Fritz and the KID: starts off: "If there is a standard opening whose reputation has been repeatedly pummeled by engines over the years, it is the King's Indian Defense. Especially the Classical main lines with the two sides rolling their pawns forward." and then goes on to say (boast?) that Fat Fritz view the KID better than other engines.
I have no point, other than to open an updated (Fall 2021) discussion on it, for folks who are more knowledgeable than I, and also to hear if there are more recent developments.
The floor is open!