correspondence chess in the age of NNUE

[Master Om]

Check this position.

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

This is from my own game. This came out from a KID opening.
Just Check the eval here. Here my opponent who is an OTB GM played 33.Qb1
This is a corr game and u can see how helpless engines are here.

Now after deep analysis I found 33...Nxg2! can be played and then Bxh3!
But check the eval and the moves suggested...

Interesting position

believe me. there is no win in sight.

This best move is only 33cp from the first 20+ moves that are all shuffling moves. Position is clearly blocked, so it is pretty basic to explore all moves that change that, and Nxg2 is the first and most promising on the list. This is nothing strange or new. Just basic stuff.

Whats basic in it is the human interpretation. But the engine part is still in flaw. There is still room for development in chess engines when it comes to a blocked position.

[Master Om]

Check this position.

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

This is from my own game. This came out from a KID opening.
Just Check the eval here. Here my opponent who is an OTB GM played 33.Qb1
This is a corr game and u can see how helpless engines are here.

Now after deep analysis I found 33...Nxg2! can be played and then Bxh3!
But check the eval and the moves suggested...

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

33... Nxg2 34. Kxg2 Bxh3+ 35. Nxh3 g4 36. fxg4 hxg4 37. Bxg4 Nxg4 38. Kh1 Rfg7
39. Nxg4 Rxg4 40. Qd1 Qh7 41. Qe2 Bh4 42. Bxh4 Rxh4 43. Rff3 Qg6 44. Ra1 Qh6
45. Raa3 Qg6 46. Ra1 Rg7 47. Qf2 Qh7 48. Raa3 Kg8 49. Qe1 Kf7 50. Qf2 Rhg4 51.
Qe3 Ke7 52. Qe1 *

If Black gives up 2 rooks for the queen Nxf4!! after doubling of rooks in f file makes the game draw.
But engines dont get it.

Now after 33...Nfg7 34.Nfg4!! hxg4 35. fxg4 the position is interesting.
See how no engines show a valid way to play this position.
After Bxh4 and Bf3 there is no way for black to penetrate white's camp.

here if any moves are played other than Ng6, white plays Bxh4 and then Bf3 closing the position. Many engines suggest to play gxh4?? which is horrible.

I don't find engine play improper here, only that the score of the position does not reflect the draw. What was improper, was steering into this position thinking it was winning. After examining this position, I see no plan to win. Checking with multiple engines will give multiple plans which are all perfectly fine to secure the draw. Engines cannot make wine from water. No engine finds a win for black because there is no win. Yes, as a correspondence player having a bogus eval is bad and leads you down the wrong path, but checking the PVs of all the suggested lines shows no progress from any engine. The super high score reflects positional advantages that appear to exist without the horizon to prove that they do not. Current engines do know how to play this position and do so well. They play it to a draw as it is a drawn position. They show an improperly large score because the 50 move rule and the repetitions are so far into the horizon it is incalculable. In the future new software might recognize this position as drawn, but in the here and now, current engines play this position flawlessly.

When the eval is with flaw how will they play flawlessly ? Both engines are blind which will play so result will also b a blind one.
also reaching this position is a standard maneuver in classical KID.

Interesting is that after 33...Rfg7 ( with the plan to play g4) my opponent played 34. Nfg4!! and after hxg4 fxg4 the position is still a draw and programs still miss-evaluate it.
[d]6rk/pp1bb1r1/3p1n1q/PPpPp1p1/2P1PpPn/R6P/4B1PN/1Q2BRK1 b - - 0 35

[Hamster]

Any random guy of 2100 using NNUE at ICCF is much more difficult to win than 6 months ago.

This sums it up very well, sadly.

[jefk]

"sums it up"

yep, although it makes it a difference whether you're
White or Black.

As for 'sadly' , well we can't hold (AI) progress, the terminators are not
yet coming anyway soon (although they can dance in Boston apparenty
and chess is still an awesome game for otb humans vs humans,

Meanwhile i think some 'crazy old guys" (as at least i once was referred
to as such in this forum (when talking about CB ) will find a solution to
make correspondence chess 'working' again (think of the endgame point
system as in Shogi, it's a possible better solution than only pieceup-> win
but then again, i also (again) think we also should compensate the (slight(*) )
advantage for White, although it are different issues indeed as sir
LK already also suggested.

Meanwhile, similar as for my rant about Cb and Chessok, i'll
be happy to include ICCF, which is part of Fide btw..
best regards
jef

(*) slight in otb chess maybe, but huge in top corresp chess

PS again, i think it's a matter of communication, ICCF, Nickel,
Kaufman, Fide, whatever, do they mail each other i wonder ,
or maybe i'm just to eager to fix it while we still have time

PS2 in the testgame i'm playing against MonteCarlo (in other thread), i
will now need to use some more time (Nf3 or g3 for those interested).
Meanwhile, considering MC's fast comp, this comp with whatever
program, will have good chances to draw with current rules.
Yet i'm still curious about the sort of endgame we will get, i
have an advantage, and it gradually will increase a little bit
with -almost- perfect play, but then not good enough for a win.
Think of BishopRookpawn vs Rook Knight or whatever, which
in the end could be an egtb draw, depending on position,
although i could be up a pawn. When counting material
(as i said above, as sometimes in Shogi endgame they do
sometimes apparently) then i might still deserve a win
with new rules. (again we then have to compensate the first move
advantage, either with time or otherwise
For more interesting reading (and the chess solving 'problem') see
https://en.wikipedia.org/wiki/First-mov ... e_in_chess

[Uri Blass]

Check this position.

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

This is from my own game. This came out from a KID opening.
Just Check the eval here. Here my opponent who is an OTB GM played 33.Qb1
This is a corr game and u can see how helpless engines are here.

Now after deep analysis I found 33...Nxg2! can be played and then Bxh3!
But check the eval and the moves suggested...

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

33... Nxg2 34. Kxg2 Bxh3+ 35. Nxh3 g4 36. fxg4 hxg4 37. Bxg4 Nxg4 38. Kh1 Rfg7
39. Nxg4 Rxg4 40. Qd1 Qh7 41. Qe2 Bh4 42. Bxh4 Rxh4 43. Rff3 Qg6 44. Ra1 Qh6
45. Raa3 Qg6 46. Ra1 Rg7 47. Qf2 Qh7 48. Raa3 Kg8 49. Qe1 Kf7 50. Qf2 Rhg4 51.
Qe3 Ke7 52. Qe1 *

If Black gives up 2 rooks for the queen Nxf4!! after doubling of rooks in f file makes the game draw.
But engines dont get it.

Now after 33...Nfg7 34.Nfg4!! hxg4 35. fxg4 the position is interesting.
See how no engines show a valid way to play this position.
After Bxh4 and Bf3 there is no way for black to penetrate white's camp.

here if any moves are played other than Ng6, white plays Bxh4 and then Bf3 closing the position. Many engines suggest to play gxh4?? which is horrible.

I don't find engine play improper here, only that the score of the position does not reflect the draw. What was improper, was steering into this position thinking it was winning. After examining this position, I see no plan to win. Checking with multiple engines will give multiple plans which are all perfectly fine to secure the draw. Engines cannot make wine from water. No engine finds a win for black because there is no win. Yes, as a correspondence player having a bogus eval is bad and leads you down the wrong path, but checking the PVs of all the suggested lines shows no progress from any engine. The super high score reflects positional advantages that appear to exist without the horizon to prove that they do not. Current engines do know how to play this position and do so well. They play it to a draw as it is a drawn position. They show an improperly large score because the 50 move rule and the repetitions are so far into the horizon it is incalculable. In the future new software might recognize this position as drawn, but in the here and now, current engines play this position flawlessly.

When the eval is with flaw how will they play flawlessly ? Both engines are blind which will play so result will also b a blind one.
also reaching this position is a standard maneuver in classical KID.

Interesting is that after 33...Rfg7 ( with the plan to play g4) my opponent played 34. Nfg4!! and after hxg4 fxg4 the position is still a draw and programs still miss-evaluate it.
[d]6rk/pp1bb1r1/3p1n1q/PPpPp1p1/2P1PpPn/R6P/4B1PN/1Q2BRK1 b - - 0 35

It is easy to solve the evaluation problem.
simply tell the engine to use smarter evaluation that is based on the result of a game between the program with stupid evaluation and itself.

Let call the engine that use the stupid evaluation engine A.

Now engine B evaluate the position based on a game between engine A and itself(when the game is at 0.01 second per move)
If after 34.Nfg4 the game end in a draw then engine B evaluate Ng4 as a draw.
If after 34.Nfg4 black win then engine B evaluate it as a forced mate for black with the number of moves that happened in the game.

Engine B is going to evaluate a fortress as a draw assuming engine A does not blunder in the fortress position during the game.
Of course mate evaluation may be wrong but evaluation is only an estimate and today also with normal chess engines 0.00 evaluation may be wrong.

Probably Engine B is going to be weaker than Engine A but I would like to see engines give the user the option to analyze with engine B and not with engine A.

Of course more time can help not only engine A but also engine B because engine B can search some plies forward and evaluate the leaves positions based on games between A and itself.

[Milos]

Check this position.

[d]6rk/pp1bbr2/3p1n1q/PPpPp1pp/2P1Pp1n/R4P1P/4BNPN/1Q2BRK1 b - - 0 33

This is from my own game. This came out from a KID opening.
Just Check the eval here. Here my opponent who is an OTB GM played 33.Qb1
This is a corr game and u can see how helpless engines are here.

Now after deep analysis I found 33...Nxg2! can be played and then Bxh3!
But check the eval and the moves suggested...

Interesting position

believe me. there is no win in sight.

This best move is only 33cp from the first 20+ moves that are all shuffling moves. Position is clearly blocked, so it is pretty basic to explore all moves that change that, and Nxg2 is the first and most promising on the list. This is nothing strange or new. Just basic stuff.

Whats basic in it is the human interpretation. But the engine part is still in flaw. There is still room for development in chess engines when it comes to a blocked position.

Basic in the sense that anyone with better than average knowledge of engine usage should be able to find it. Which should be any CC player with rating above 1800. Problem is that most of CC players are really bed.
Engines have their limitations, and I don't expect things to change much in this direction, because it's simply too difficult to reliably detect certain type of positions (like in the example above) and not lose Elo. But that is less than a few % of positions, and that's where human intervention can actually produce better result than stand alone engine.

[mbabigian]

"sums it up"

yep, although it makes it a difference whether you're
White or Black.

As for 'sadly' , well we can't hold (AI) progress, the terminators are not
yet coming anyway soon (although they can dance in Boston apparenty
and chess is still an awesome game for otb humans vs humans,

Meanwhile i think some 'crazy old guys" (as at least i once was referred
to as such in this forum (when talking about CB ) will find a solution to
make correspondence chess 'working' again (think of the endgame point
system as in Shogi, it's a possible better solution than only pieceup-> win
but then again, i also (again) think we also should compensate the (slight(*) )
advantage for White, although it are different issues indeed as sir
LK already also suggested.

Meanwhile, similar as for my rant about Cb and Chessok, i'll
be happy to include ICCF, which is part of Fide btw..
best regards
jef

(*) slight in otb chess maybe, but huge in top corresp chess

PS again, i think it's a matter of communication, ICCF, Nickel,
Kaufman, Fide, whatever, do they mail each other i wonder ,
or maybe i'm just to eager to fix it while we still have time

PS2 in the testgame i'm playing against MonteCarlo (in other thread), i
will now need to use some more time (Nf3 or g3 for those interested).
Meanwhile, considering MC's fast comp, this comp with whatever
program, will have good chances to draw with current rules.
Yet i'm still curious about the sort of endgame we will get, i
have an advantage, and it gradually will increase a little bit
with -almost- perfect play, but then not good enough for a win.
Think of BishopRookpawn vs Rook Knight or whatever, which
in the end could be an egtb draw, depending on position,
although i could be up a pawn. When counting material
(as i said above, as sometimes in Shogi endgame they do
sometimes apparently) then i might still deserve a win
with new rules. (again we then have to compensate the first move
advantage, either with time or otherwise
For more interesting reading (and the chess solving 'problem') see
https://en.wikipedia.org/wiki/First-mov ... e_in_chess

You keep saying Montecarlo has a fast computer. Repeatedly saying it gives you an out to save face, but doesn't make it true. My wife's machine which needs replacing is about as fast as his. (I'm in the process of trying to find parts now) The tablet I'm typing this on does 2.5 million nps for heaven sake.

My computer is fast, at 78 million to well over 100 million nps depending on the position and software.

I think your computing frame of reference is disconnected from reality.

[jefk]

Mike ('mbabigian') wrote:

My computer is fast, at 78 million to well over 100 million nps depending on the position and software.

yes that's fast, faster than MC's comp indeed; but
for normal chess playing online as human you don't
such a monster, you know...
That i mentioned MC comps' speed apparently more than once
(although there now are 2 threads, so that may have been a reason)
is not to 'save *my* face' btw, but apparently if SF doesn't see a
move (due it's eval) then it often will also not see such a move
if you calculate deeper. Which explains i'm already
in an advantage (*) in the test game were' now playing
(whether that'ts enough to win the game, remains to be seen..

I think your computing frame of reference is disconnected from reality.

You can think what you want but you don't know me; although
i suspected Moore;s law was slowing down years ago,
this apparently has not happened (yet). For the rest, fyi
(after a speculation which had gone wrong) i had some
bad luck in recent years with the Dutch tax inquisition,
and thus saved on computer costs (yes some 2nd hand
servers (as Ed S once mentioned) are an option, but then after
one/1 years they may be outdated again as well; besides that
for me it's also a challenge to play correspondence chess
without an extremely fast machine (i once got the tip
from Bas Hamstra that good programmers don't need fast
comps, in fact the best software is written on simple comps
(with eg. Linux or so, although that's another topic).

PS (*) hesitated whether i should give move comments, about
my current game but anyway, even with my apparently nowadays
slow machine i'm now already in an advantage, due to
MC's move 1...Nf6? instead of 1...d5! Reason, after 1...Nf6
(as in current game) and than 2.c4 e6, i can play 3.g3, with
the awesome option of playing 4.Nbd2!!! after 3...Bb4+
(instead of 4.Bd2 (*****), if d5 would already have been played
(Nd2
then also is possible, but not as good).
LK recommends the Gruenfeld in his latest book, and after f3
the strange move ...Nc6? as recommended in a book by Agrest etc al.
However after a later d5 Ne5 etc (which LK not even mentions
in his book), White maintains an advantage. Doesn't seem
important but it turns out to be crucial in the decision to
start with 1.d4 vs 1.e4 (i didnt' go for 1.Nf3 because after 1...c5 2.c4
the symmetrical English, White has an advantage, although after variations
with a later ...g6 it may be very hard to win the game.
Maybe, while playing this test game (in the other thread), i'll
draft a new book with possible title 'winning with 1.d4) or so (****).
Anyway i expect within some ten moves also, to come up with
an interesting novelty (at least vs GM play), and imho it will
become an interesting game. As for my 'face' ie reputation, i
don't give a damn btw, some people like my work, others don't.
eg. i have been sneered at sometimes because i'm not an Otb
master player, but this is simply because i'm a lazy otb player,
and have other interests; in my latest weekend tourn in Leiden
in otb i scored above 1800, and on Lichess with rapid i'm also
above 1800, with my ICCF rating 2330 or so (with my 'slow'
comp, indeed). My otb rating doesn't reflect my knowledge
about chess inmho; the latter i approach in a scientific way,
and i have quite some middlegame/endgame books which i
can look at while playing a corresp game. Besides i'll becoming
66 yrs next week, and whereas i started with serious chess (at a
chess club) at the relatively late age of 35, i now have no intention
of competing with young little pricks anymore to become eg a Fide
master in otb chess. Once handicaps have been introduced in otb
chess (as i propose, similar as in Shogi) it may become more interesting,
and less antisocial btw and i may give it a go again. Antisocial ? yes
antisocial (as also mentioned once here by HGM), in the chess language for
example they often talk about torturing your opponent (eg in
the RL) and stuff like that. A GM as Simon Williams wrote a
book (maybe tongue in cheek referring to RJ Fischer) 'how to
crush your opponent' (it's not such a bad book btw, and contains
decent chess suggestions, but anyway). If your fast comp makes you
feel better to compensate for other inadequacies (eg in chess),that's
fine with me. But (tip for you (**)) don't suggest stuff as eg that i'm
'disconnected from reality' (***) ...
(**) fyi i tend to be a good debater in online forums, believe it or not,
and if you continue like this, you may well try to save your own face
before the dispute is ended (or a moderator takes action).
(***) when i wrote many years ago that chess is a draw (more than
a century later than Steinitz btw,) i got lots of aggressive reactions.
Now that i suggest White still has an advantage (ie that in the
remaining future non-drawn games, most will be won with White),
again i get some aggressive reactions... Well i got used to it, you know.
(****) or, which would be more fun (and probably would sell better,
a title 'The crazy world of computer chess' or so; after

")
PS fortunately not all gifted young players are little psycho's (and i also hope for you, you also don't fall into that category btw), eg. a certain Magnus C
(yep met him once) is a nice guy, although tough indeed in
endgame play etc. The Dutch (former superGM Timman was
a normal guy (and still is, a few years older than me btw.
(****) even with Bd2 (normal Bogo-Indian White achieves the
bishop pair, and thus a better endgame (something which is my
purpose to demonstrate, and which is relevant once endgame
rules would be changed. Although the suggestion to reduce
White's time to eg. one day to compensate, i still don't abandon
my idea of a default starting handicap (remove f2 and c7 at the
start) for correspondence chess, because the new NNu comps/programs
can easily handle such position with ease, without opening book.
Maybe another test game later to explore this would also be fun;
but yep, then i probably will have upgraded my current comp (lol)....

[Master Om]

Any random guy of 2100 using NNUE at ICCF is much more difficult to win than 6 months ago.

This sums it up very well, sadly.

I am agreemententing...

[mbabigian]

Mike ('mbabigian') wrote:

My computer is fast, at 78 million to well over 100 million nps depending on the position and software.

yes that's fast, faster than MC's comp indeed; but
for normal chess playing online as human you don't
such a monster, you know...
That i mentioned MC comps' speed apparently more than once
(although there now are 2 threads, so that may have been a reason)
is not to 'save *my* face' btw, but apparently if SF doesn't see a
move (due it's eval) then it often will also not see such a move
if you calculate deeper. Which explains i'm already
in an advantage (*) in the test game were' now playing
(whether that'ts enough to win the game, remains to be seen..

I think your computing frame of reference is disconnected from reality.

You can think what you want but you don't know me; although
i suspected Moore;s law was slowing down years ago,
this apparently has not happened (yet). For the rest, fyi
(after a speculation which had gone wrong) i had some
bad luck in recent years with the Dutch tax inquisition,
and thus saved on computer costs (yes some 2nd hand
servers (as Ed S once mentioned) are an option, but then after
one/1 years they may be outdated again as well; besides that
for me it's also a challenge to play correspondence chess
without an extremely fast machine (i once got the tip
from Bas Hamstra that good programmers don't need fast
comps, in fact the best software is written on simple comps
(with eg. Linux or so, although that's another topic).

PS (*) hesitated whether i should give move comments, about
my current game but anyway, even with my apparently nowadays
slow machine i'm now already in an advantage, due to
MC's move 1...Nf6? instead of 1...d5! Reason, after 1...Nf6
(as in current game) and than 2.c4 e6, i can play 3.g3, with
the awesome option of playing 4.Nbd2!!! after 3...Bb4+
(instead of 4.Bd2 (*****), if d5 would already have been played
(Nd2
then also is possible, but not as good).
LK recommends the Gruenfeld in his latest book, and after f3
the strange move ...Nc6? as recommended in a book by Agrest etc al.
However after a later d5 Ne5 etc (which LK not even mentions
in his book), White maintains an advantage. Doesn't seem
important but it turns out to be crucial in the decision to
start with 1.d4 vs 1.e4 (i didnt' go for 1.Nf3 because after 1...c5 2.c4
the symmetrical English, White has an advantage, although after variations
with a later ...g6 it may be very hard to win the game.
Maybe, while playing this test game (in the other thread), i'll
draft a new book with possible title 'winning with 1.d4) or so (****).
Anyway i expect within some ten moves also, to come up with
an interesting novelty (at least vs GM play), and imho it will
become an interesting game. As for my 'face' ie reputation, i
don't give a damn btw, some people like my work, others don't.
eg. i have been sneered at sometimes because i'm not an Otb
master player, but this is simply because i'm a lazy otb player,
and have other interests; in my latest weekend tourn in Leiden
in otb i scored above 1800, and on Lichess with rapid i'm also
above 1800, with my ICCF rating 2330 or so (with my 'slow'
comp, indeed). My otb rating doesn't reflect my knowledge
about chess inmho; the latter i approach in a scientific way,
and i have quite some middlegame/endgame books which i
can look at while playing a corresp game. Besides i'll becoming
66 yrs next week, and whereas i started with serious chess (at a
chess club) at the relatively late age of 35, i now have no intention
of competing with young little pricks anymore to become eg a Fide
master in otb chess. Once handicaps have been introduced in otb
chess (as i propose, similar as in Shogi) it may become more interesting,
and less antisocial btw and i may give it a go again. Antisocial ? yes
antisocial (as also mentioned once here by HGM), in the chess language for
example they often talk about torturing your opponent (eg in
the RL) and stuff like that. A GM as Simon Williams wrote a
book (maybe tongue in cheek referring to RJ Fischer) 'how to
crush your opponent' (it's not such a bad book btw, and contains
decent chess suggestions, but anyway). If your fast comp makes you
feel better to compensate for other inadequacies (eg in chess),that's
fine with me. But (tip for you (**)) don't suggest stuff as eg that i'm
'disconnected from reality' (***) ...
(**) fyi i tend to be a good debater in online forums, believe it or not,
and if you continue like this, you may well try to save your own face
before the dispute is ended (or a moderator takes action).
(***) when i wrote many years ago that chess is a draw (more than
a century later than Steinitz btw,) i got lots of aggressive reactions.
Now that i suggest White still has an advantage (ie that in the
remaining future non-drawn games, most will be won with White),
again i get some aggressive reactions... Well i got used to it, you know.
(****) or, which would be more fun (and probably would sell better,
a title 'The crazy world of computer chess' or so; after

")
PS fortunately not all gifted young players are little psycho's (and i also hope for you, you also don't fall into that category btw), eg. a certain Magnus C
(yep met him once) is a nice guy, although tough indeed in
endgame play etc. The Dutch (former superGM Timman was
a normal guy (and still is, a few years older than me btw.
(****) even with Bd2 (normal Bogo-Indian White achieves the
bishop pair, and thus a better endgame (something which is my
purpose to demonstrate, and which is relevant once endgame
rules would be changed. Although the suggestion to reduce
White's time to eg. one day to compensate, i still don't abandon
my idea of a default starting handicap (remove f2 and c7 at the
start) for correspondence chess, because the new NNu comps/programs
can easily handle such position with ease, without opening book.
Maybe another test game later to explore this would also be fun;
but yep, then i probably will have upgraded my current comp (lol)....

Jefk, thanks for the wall of text - relax, LOL. Just out of curiosity I downloaded the latest SF-dev on my wife's machine. Without making a machine specific compile or bothering to enable large pages, I got 9.7M nps. The processor is 7 years old and the OS and CPU are so full of unpatchable security holes, it is no longer safe to use. With a custom compile and large pages it would have shattered 10M nps.

If you consider a 7 year old machine fast, you are in a small minority. Best of luck avoiding the draw against MonteCarlo and his "fast" machine. It likely won't be easy.

Peace.