lichess.org

Why is the Engine move in the Italian Game vs Two Knights Defense different from it's color mirror?

6 d4 used to be first choice until 6...Bb4+ 7 c3 Be7 8 Nxf7+ Kxf7 9 Qf3+ Ke6 was discovered, where white cannot play Nc3.
Since then 6 Nxf7 is first choice.
Back on topic: the engine also stores possible moves and analyses as the game progresses. So after 1 e4 or after the same position with e3 and e4 with colors reversed, the engine starts with a different storage of previously calculated lines and evaluations so it reaches a different result.
Humans are weak at openings, but humanity is not. Human opening knowledge is derived from millions of played games, i.e. millions of hours of concentrated thought and even more analysis wthout and with engines. An engine cannot match that in a few minutes of calculation.
The engine does not even look at all lines, it only looks at a few pruned lines, as per its previous conclusions.

Engines do look all the lines. Just thet prove un-necessary lines i.e if move loses at least a pawn no poit calculating if loses a bishop if neutral move has already found. well null-move prunin can remove some not bad lines but rarely. Neural net engines look only interesting lines but that whole different strory. And Nope computer does not keep in memory how we got here it looks at situation as humas should. Caring only what on board and how it came to be.

Actually, Leela Chess Zero needs the game history really badly. I'm not sure about newest versions, but older ones used to lose a few hundred Elo points when given a position without previous moves. If in some computer chess tournaments games are played with odds, Leela will need artificial game history up to the starting position.

Leela in a neural network machine so it very different from other engined. Though I really dont see why woudl it need history?

Any insight into the neural network mechanism is always very difficult. It was trained with game history and it somehow "got used to it", but I doubt whether even developers would try to describe the exact reason. Maybe, just maybe, it is able to derive some information about its opponent's strategy?

i've always assumed all 'standard' engines (the usual suspects) have opening 'books' stored in them that they use for openings, and that that is the typical way the engine would approach the opening. i got this impressions reading a book - i think it was the one written by the designer of deep blue,,, or was it deep mind... the computer that beat kasparov.

so --- i guess i'd assume computers of that generation go by the book, roughly, first and foremost, not endlessly going through non-book moves. i've assumed they're much stronger than any human in an opening.

i have no idea how neural network computers calculate.... they're pretty interesting in theory, tho, and i heard they are stronger than their predecessors. there's a book about them too: game changer.

An engine do not use an opening book unless you set it up to. Modern engines can play decently in the opening without them, but of course a good opening book still helps for strength as you can have thousands of CPU hours reused in a blink by looking in the book.

Lichess's Stockfish doesn't use any opening book however.

NN engines behave differently, their training method puts a big emphasis on memorizing the best way to play in early positions, so the opening phase is their biggest strength usually.

@Alayan i'm p surprised modern engines don't use opening 'books.' in a sense, wouldn't that give the human a sizable advantage? i mean, that guy who spent half of his life studying the ruy lopez and knowing every variation... it seems like, yeah, the engine would win (sorry guy) but it'd spend billions of cycles going thrue idiotic possibilities... not that its silicon heart gives a uh hoot.

i've only 'heard' about the NN engines -------- uh,,... that they tell them the rules of the game, and that's it! then pit them against each other in an iron cage for awhile.... and then, voila! a chess savant is born! i guess i'm under the impression we don't understand how they think and strategize until we analyze the game itself and somehow come to our own conclusions... and that we don't train them in any way. they train themselves by playing countless games against each other -

sorry if i sound like a simpleton here.........

NN engines do not memorize anything per se. They just adjust weights to reflect outcomes of the games in training. And with those weights they do come into conlcusions as what they think is best opening.

@doublebanzai engines dont have opening book as it cleare solution let program above to decide what opening book to use. engine only for parts that need engine

You can't post in the forums yet. Play some games!

Reconnecting