Best engine evaluation leads to white loss in French Defense line
Sort:
[COMMENT DELETED]
n8sl8 wrote:
I have just started trying to learn the French defense. White has played the: Advance Variation, Nimzowitsch System (1. e4 e6 2. d4 d5 3. c4 c5 4.Nf3) on me a couple times already and I was trying to find a good response. I saw that 4...cxd4 was a possibility. I then put that line into the analysis feature on chess.com. I picked the best line it would come up with to see how the game might play out (it went up to move 12 or so). I decided to just keep going by choosing the best continuing line after that, and went all the way until the game was finished which ended in a rather dominant victory for black. I certainly don't think that position after 4 moves that I put into the analysis is doomed for white by any means. Why doesn't Stockfish - proven this week to be the best chess engine in blitz - come up with a line that ends in a draw? How can the engine beat itself? (My settings were Lines: 7, Engine: Stockfish.js 8, Time limit: 10 seconds - which I know is low, but still, this is blitz and it's a supercomputer).
You forgot to advance. Where's 3 e5?
3 c4 is not the move. It's not the French Advance.
It's wrong, bad and evidently losing.
Nov 18, 2017
0
#4
notmtwain: Sorry, 3. e5 is what I meant to write (obviously 3. c4 is not that good). I corrected it in the post. I was giving the line for French Defense: Advance Variation, Nimzowitsch System which people have played against me.
So, what did you play after 1.e4 e6 2.d4 d5 3.e5 c5 4.Nf3 cxd4?
If you picked any of the two top suggestions of the engines (5.Nxd4 or 5.Qxd4) then it is natural to follow into an unpleasant position. Nimzowitz himself played them in few occasions, but eventually gave up on them.
White's only consistent plan positionally is 5.Bd3! which the engines do not evaluate that high, but this is natural: Engines just calculate, they do not PLAN.
3. e5 is still not a bad move, but much worse than 3. Nc3, so one suggestion is to simply switch to that.
Even better is to play 2. d3, and then switch to a KIA.
In the last couple of years, I have not won a single pawn ending against a top engine,
and maybe just one or 2 minor piece endings.
Top engines are unbeatable in the endgame these days, you should try your luck
earlier on, when the game is more complicated.
Nov 20, 2017
0
#9
n8sl8 wrote:
... How can the engine beat itself? (My settings were Lines: 7, Engine: Stockfish.js 8, Time limit: 10 seconds - which I know is low, but still, this is blitz and it's a supercomputer).
The chess.com engine is Stockfish but is running 100% on the client side and is limited by your computer's processing power; it certainly isn't a super computer. Add that to the 10 seconds and you can get weird evaluations, especially in complex positions.
Not going to speak at all to the line, since I don't know anything about it or best play, there.
pfren wrote:
White's only consistent plan positionally is 5.Bd3! which the engines do not evaluate that high, but this is natural: Engines just calculate, they do not PLAN.
Interesting....after 5 bd3 nc6 6 o-o a6, it had black at a .4 advantage, with a recommendation of 7c3 or Bf4 bringing black to a .5 advantage. But when I entered 7 re1, it now said black was at a .17 advantage only. It gets better...when I clicked back on 6...a6, it said 7 re1 was the best move at a .10 advantage for black. I went from scratch and put the moves in and on move 7 it says 7re1 is a .33 advantage for black...so I played 7re1 and it said .14 advantage for black. I clicked on the last move, and it still had black at a .14 advantage(with 27 moves completed). One more go round with 36 moves completed had black at a .04 advantage. Of course when I clicked back to move 27, it gave a different move for white on 28 and a .25 advantage for black.
Edit: I should have said that white still moved 8.c3 in that scenario, giving up a pawn. After 7....Qc7, I went with Nbd2 and it said .28 advantage black. Moving the game forward to move 22 it now said white had the advantage.
Let the show go on.
The good thing now is Deirdre Sky is my best friend. 
Time matters for some positions, but not for others.
There are an awful lot of positions, where, even if you give Stockfish centuries,
it will still output the same wrong output.
Top engines prune a lot, the stronger they are, the more they prune, so some moves, shallow
in the tree, are not considered at all. If you don't consider some moves shallow in the tree,
then you are pruning much more later on, so it is basically a mess.
Top engines are statistical creatures, they know how to capture and give check, everything else
is just stats.
I guess with humans it is more or less the same.
If you have followed Talkchess, there we have refuted almost every single theory and great human move ever played.
For example, it proved Kasparov and Deep Blue have been blundering on each and every move.
Let's take only this famous position from 1997:
The whole dispute here is whether Deep Blue blundered with 45. Ra6, where Kasparov could possibly have drawn with Qe3 to give a (probable) perpetual.
Some argue Ra6 missed the win, others hold the opposite opinion, we have been analysing this for some 30 or so moves forward, and there is still no definite conclusion if Ra6 misses the win or not.
Too complicated.
Top engines nowadays see Qd7 instead of Ra6 is better, and should win, but they still don't see the best move, namely Qb6.
Qb6 has been also missed by all commentators, very funny.
So why should I rely on Stockfish and Komodo, when they think Qd7 is best, or follow those
unanimous commentators of GM strength some of them, who blidly trust the engines and still don't even consider Qb6?
Considering evaluation, evaluation is always static, the sum total of all available chess knowledge patterns on the board, so it can never have anything in common with a plan.
You evaluate the separate chess positions along the tree, and there is no way you can include those in a plan.
DeirdreSkye wrote:
Qb6 is not possible.Someone who has spend so much time analysing the position would know that(can't be a typo since the same mistake is repeated 3 times).
Do you mean Qxb6?
You can't be that pedantic, can you?
This is not a pgn.
There is short and long algebraic notation, epd description, etc.
We must just exercise some common sense here.
Another example, above diagram is from Carlsen-Aronyan.
I don't know how many sites and videos have referred to Aronian's a3 above as a great move, attaching lots of exclams, when it is actually more of a blunder.
Deeper analysis shows that a3 leads only to a draw, while c5 instead is much better.
So, Aronyan actually blundered and everybody is giving him exclams on all possible sites and videos.
Weaker players, the majority of the chess population, seeing this undisputable evidence, are quick to
applaud, so everybody accepts without reserve a3 is best and deserves applause.
Why should I care too much about such sites, such videos and such players?
nighteyes1234 έγραψε:
pfren wrote:
White's only consistent plan positionally is 5.Bd3! which the engines do not evaluate that high, but this is natural: Engines just calculate, they do not PLAN.
Interesting....after 5 bd3 nc6 6 o-o a6, it had black at a .4 advantage, with a recommendation of 7c3 or Bf4 bringing black to a .5 advantage. But when I entered 7 re1, it now said black was at a .17 advantage only. It gets better...when I clicked back on 6...a6, it said 7 re1 was the best move at a .10 advantage for black. I went from scratch and put the moves in and on move 7 it says 7re1 is a .33 advantage for black...so I played 7re1 and it said .14 advantage for black. I clicked on the last move, and it still had black at a .14 advantage(with 27 moves completed). One more go round with 36 moves completed had black at a .04 advantage. Of course when I clicked back to move 27, it gave a different move for white on 28 and a .25 advantage for black.
Edit: I should have said that white still moved 8.c3 in that scenario, giving up a pawn. After 7....Qc7, I went with Nbd2 and it said .28 advantage black. Moving the game forward to move 22 it now said white had the advantage.
Oh boy... no, not like that.
Drop your engine to the wastebin: it does a lot of harm to your thinking-actually it does not allow you to think at all, which is the worst thing you would expect when working to improve.
Get a good book on the French instead.
Nov 21, 2017
0
#16
Get a good book on the French instead.
French what ?....revolution, kiss, toast ?....what ?
The French Advanced Kiss, if you happen to know what that is.
When I think now I spent week and a half analysing Aronyan-Carlsen this summer...
Running statistically relevant engine gauntlets, top engine shootouts, checking manually different lines
with engine help, etc.
When I think now, it is much better to just play bullet games.
That is what I am saying, engines are very weak.
I have just started trying to learn the French defense. White has played the: Advance Variation, Nimzowitsch System (1. e4 e6 2. d4 d5 3. e5 c5 4.Nf3) on me a couple times already and I was trying to find a good response. I saw that 4...cxd4 was a possibility. I then put that line into the analysis feature on chess.com. I picked the best line it would come up with to see how the game might play out (it went up to move 12 or so). I decided to just keep going by choosing the best continuing line after that, and went all the way until the game was finished which ended in a rather dominant victory for black. I certainly don't think that position after 4 moves that I put into the analysis is doomed for white by any means. Why doesn't Stockfish - proven this week to be the best chess engine in blitz - come up with a line that ends in a draw? How can the engine beat itself? (My settings were Lines: 7, Engine: Stockfish.js 8, Time limit: 10 seconds - which I know is low, but still, this is blitz and it's a supercomputer).