Whatever is current - Amazing Leela

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
corres
Posts: 1770
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Whatever is current - Amazing Leela

Post by corres » Fri Jul 12, 2019 7:32 am

mclane wrote:
Thu Jul 11, 2019 8:51 pm
Yes. It’s a new paradigm.
And like usual, the mass is not seeing the diamonds in front of them.
Na ja!
The "new paradigm" can be seen for "smart" only.
With this the "smart" is smart, is not it?

Dann Corbit
Posts: 10198
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Whatever is current - Amazing Leela

Post by Dann Corbit » Fri Jul 12, 2019 7:48 am

corres wrote:
Fri Jul 12, 2019 7:23 am
Dann Corbit wrote:
Fri Jul 12, 2019 6:29 am
All the same, there is a point of interest here.
We do not onlly wish to understand the result.
We wish (perhaps even more so) to understand the path to the result.
Most of the time, I know by statistics which move is best for a given position. So when I analyze it, I do not care about what but instead why.
I agree but the statistics may change sometimes drastically when a variation was subverted.
Until game outcome is determined everything is a guess (estimate/extrapolation/approximation).
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

corres
Posts: 1770
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Whatever is current - Amazing Leela

Post by corres » Fri Jul 12, 2019 8:08 am

Dann Corbit wrote:
Fri Jul 12, 2019 7:48 am
corres wrote:
Fri Jul 12, 2019 7:23 am
Dann Corbit wrote:
Fri Jul 12, 2019 6:29 am
All the same, there is a point of interest here.
We do not onlly wish to understand the result.
We wish (perhaps even more so) to understand the path to the result.
Most of the time, I know by statistics which move is best for a given position. So when I analyze it, I do not care about what but instead why.
I agree but the statistics may change sometimes drastically when a variation was subverted.
Until game outcome is determined everything is a guess (estimate/extrapolation/approximation).
O.K. But it is important for an exercising player how high those numbers are.

todd
Posts: 52
Joined: Thu Apr 19, 2018 7:09 pm

Re: Whatever is current - Amazing Leela

Post by todd » Fri Jul 12, 2019 5:17 pm

Ovyron wrote:
Thu Jul 11, 2019 10:59 pm
I think that falls outside of the scope of this thread (where I've been claiming you should focus on the relevant chess lines instead of generic chess lines, so if it's "not played as often in engine tournaments" that means the line is generic), but, sure, I'll see what I can do about those lines and we can discuss them in a new thread that I'll create.
Fair enough. Indeed it is important to focus on the strengths and weaknesses of yourself and your opponents.

And if you are a computer playing against other computers, in the Italian, I guess if you want to score well with black, you goad your opponents into playing a Bg5 Giuoco. That's not all you do, and humans playing humans should do different things because their own strengths and weaknesses are different, but I guess you're right that this is a bit off topic and you probably don't want too much information shared on other lines where white makes mistakes in engine play, so I'll cut this short.

User avatar
Ovyron
Posts: 2826
Joined: Tue Jul 03, 2007 2:30 am

Re: Whatever is current - Amazing Leela

Post by Ovyron » Sun Jul 14, 2019 6:50 am

todd wrote:
Fri Jul 12, 2019 5:17 pm
Ovyron wrote:
Thu Jul 11, 2019 10:59 pm
I think that falls outside of the scope of this thread (where I've been claiming you should focus on the relevant chess lines instead of generic chess lines, so if it's "not played as often in engine tournaments" that means the line is generic), but, sure, I'll see what I can do about those lines and we can discuss them in a new thread that I'll create.
Fair enough. Indeed it is important to focus on the strengths and weaknesses of yourself and your opponents.
Sorry. okay, no.

While my backsolved tree graph has a score that is positive for the black side after Ng5, it turns out to be an obscure line, and in practice people don't know how to play against the Traxler... So no way I'm going to put it out in the open so the playing field is leveled, not if my only chance to contend against faster hardware and NNs (I can't run GPU Leela and CPU Leela is cute but hilarious) is powerful uncorked lines, and specially, the person that gave me my private book made me promise that I'd not share it with anyone (they were so worried, they checked back to make sure I understood the implications), so what's the difference if I just go and divulge these lines in public?

But, yes, I'll be contending with the Traxler, because, as it turns out, it's a good variation to play both with white and black, because engines misplay the black side.

I'm copping out.
todd wrote:
Fri Jul 12, 2019 5:17 pm
And if you are a computer playing against other computers, in the Italian, I guess if you want to score well with black
I don't want to score well with black, because I already score well with black. I eat my cake and have it too :mrgreen:
todd wrote:
Fri Jul 12, 2019 5:17 pm
you goad your opponents into playing a Bg5 Giuoco.
A critical line has Bg5, which means it's white's best try (specially against opponents that don't know black's critical continuation) so that other white alternatives do worse. That's why I believe this has nothing to do with how engines misplay a side, but about equal opponents that know the lines being better avoiding it all as white (1.d4 has a similar line, that's why some people just play 1.c4).
todd wrote:
Fri Jul 12, 2019 5:17 pm
That's not all you do, and humans playing humans should do different things because their own strengths and weaknesses are different
I doubt a human would ever want to play optimal black's defense in the Italian, Stockfish gives +0.50 advantage to white for a reason, it's very difficult for even top engines on fastest hardware to find the correct continuation for black, humans would just be... busted.
Great spirits have always encountered violent opposition from mediocre minds.

Uri Blass
Posts: 8606
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Whatever is current - Amazing Leela

Post by Uri Blass » Sun Jul 14, 2019 7:04 am

Ovyron wrote:
Sun Jul 14, 2019 6:50 am
todd wrote:
Fri Jul 12, 2019 5:17 pm
Ovyron wrote:
Thu Jul 11, 2019 10:59 pm
I think that falls outside of the scope of this thread (where I've been claiming you should focus on the relevant chess lines instead of generic chess lines, so if it's "not played as often in engine tournaments" that means the line is generic), but, sure, I'll see what I can do about those lines and we can discuss them in a new thread that I'll create.
Fair enough. Indeed it is important to focus on the strengths and weaknesses of yourself and your opponents.
Sorry. okay, no.

While my backsolved tree graph has a score that is positive for the black side after Ng5, it turns out to be an obscure line, and in practice people don't know how to play against the Traxler... So no way I'm going to put it out in the open so the playing field is leveled, not if my only chance to contend against faster hardware and NNs (I can't run GPU Leela and CPU Leela is cute but hilarious) is powerful uncorked lines, and specially, the person that gave me my private book made me promise that I'd not share it with anyone (they were so worried, they checked back to make sure I understood the implications), so what's the difference if I just go and divulge these lines in public?

But, yes, I'll be contending with the Traxler, because, as it turns out, it's a good variation to play both with white and black, because engines misplay the black side.

I'm copping out.
todd wrote:
Fri Jul 12, 2019 5:17 pm
And if you are a computer playing against other computers, in the Italian, I guess if you want to score well with black
I don't want to score well with black, because I already score well with black. I eat my cake and have it too :mrgreen:
todd wrote:
Fri Jul 12, 2019 5:17 pm
you goad your opponents into playing a Bg5 Giuoco.
A critical line has Bg5, which means it's white's best try (specially against opponents that don't know black's critical continuation) so that other white alternatives do worse. That's why I believe this has nothing to do with how engines misplay a side, but about equal opponents that know the lines being better avoiding it all as white (1.d4 has a similar line, that's why some people just play 1.c4).
todd wrote:
Fri Jul 12, 2019 5:17 pm
That's not all you do, and humans playing humans should do different things because their own strengths and weaknesses are different
I doubt a human would ever want to play optimal black's defense in the Italian, Stockfish gives +0.50 advantage to white for a reason, it's very difficult for even top engines on fastest hardware to find the correct continuation for black, humans would just be... busted.
a human does not need to find the moves assuming all is home preperation so I believe 2700+ players are going to have motivation to learn the correct lines in order to win.

User avatar
Ovyron
Posts: 2826
Joined: Tue Jul 03, 2007 2:30 am

Re: Whatever is current - Amazing Leela

Post by Ovyron » Sun Jul 14, 2019 7:36 am

Uri Blass wrote:
Sun Jul 14, 2019 7:04 am
a human does not need to find the moves assuming all is home preperation so I believe 2700+ players are going to have motivation to learn the correct lines in order to win.
All this has been going on for more than a decade, for example, back when Freestyle chess was popular, those players were corking the most powerful opening lines that were seen at that point in history. People were going insane building strong clusters to run Rybka and derivatives that were stronger than it in some lines, looking for anything that would give them the edge against the competition.

What was absent was a GM looking at this, or caring (what was there were very strong human players where their chess knowledge was useless against the strongest centaurs, which was in part responsible for freestyle's death as it was won by the unknowns). The humans had their own preparation and it has nothing to do with what is best on corr chess, freestyle chess, or engine matches. Now and then the GMs will play a variation that appeared previously on some famous engine game, but this is rare.

The difference in strength is so enormous between engines and humans, that the best possible moves engines can play are useless to them, just like 1600 elo patzers would find useless to memorize and play GM lines like parrots just to lose because they didn't notice their Bishop was hanging on the next move.
Great spirits have always encountered violent opposition from mediocre minds.

Uri Blass
Posts: 8606
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Whatever is current - Amazing Leela

Post by Uri Blass » Sun Jul 14, 2019 8:33 am

Ovyron wrote:
Sun Jul 14, 2019 7:36 am
Uri Blass wrote:
Sun Jul 14, 2019 7:04 am
a human does not need to find the moves assuming all is home preperation so I believe 2700+ players are going to have motivation to learn the correct lines in order to win.
All this has been going on for more than a decade, for example, back when Freestyle chess was popular, those players were corking the most powerful opening lines that were seen at that point in history. People were going insane building strong clusters to run Rybka and derivatives that were stronger than it in some lines, looking for anything that would give them the edge against the competition.

What was absent was a GM looking at this, or caring (what was there were very strong human players where their chess knowledge was useless against the strongest centaurs, which was in part responsible for freestyle's death as it was won by the unknowns). The humans had their own preparation and it has nothing to do with what is best on corr chess, freestyle chess, or engine matches. Now and then the GMs will play a variation that appeared previously on some famous engine game, but this is rare.

The difference in strength is so enormous between engines and humans, that the best possible moves engines can play are useless to them, just like 1600 elo patzers would find useless to memorize and play GM lines like parrots just to lose because they didn't notice their Bishop was hanging on the next move.
I do not think that it is useless for humans and getting an advantage increase chances to win.

Komodo can win me with 2 pawns handicap but it does not mean that winning a pawn against human opponents is useless for me.
Winning a pawn increase significantly my chances to win against other human opponents.

Even if GM's stop to memorize in a winning position when they practically lose against computers then the position is still a better starting point for them in a game.

zullil
Posts: 5683
Joined: Mon Jan 08, 2007 11:31 pm
Location: PA USA
Full name: Louis Zulli

Re: Whatever is current - Amazing Leela

Post by zullil » Sun Jul 14, 2019 10:39 am

Ovyron wrote:
Sun Jul 14, 2019 6:50 am
is powerful uncorked lines, and specially, the person that gave me my private book made me promise that I'd not share it with anyone (they were so worried, they checked back to make sure I understood the implications), so what's the difference if I just go and divulge these lines in public?

But, yes, I'll be contending with the Traxler, because, as it turns out, it's a good variation to play both with white and black, because engines misplay the black side.

I'm copping out.
Keeping a promise is not "copping out."

Common sense tells us that 1. e4 e5 2. Nf3 Nc6 3. Bc4 is a draw with correct play. You believe you've found a line that gives you (i.e., some engine you are running) good practical chances for Black against current chess engines at 12+2 time control. Fine.

User avatar
Ovyron
Posts: 2826
Joined: Tue Jul 03, 2007 2:30 am

Re: Whatever is current - Amazing Leela

Post by Ovyron » Sun Jul 14, 2019 5:56 pm

Uri Blass wrote:
Sun Jul 14, 2019 8:33 am
Winning a pawn increase significantly my chances to win against other human opponents.
What if the best line for black includes sacrificing a pawn, and both exchanges (you keep the Bishop pair, so it's knights for rooks), so you end with huge material deficit in a position you have no idea how to continue? That's the kind of stuff I'm seeing, if GMs hire people to keep them up to date about latest opening theory, and you don't see these lines at top level play, it's because these lines are useless at the human level, because they're way too complex and it's hard to know what's going on (even engines get confused.) Top GMs are best playing into positions that they know how to play, and it wouldn't matter to them how Stockfish or Leela plays them. The best positions against engines are not the same as against humans.
zullil wrote:Common sense tells us that 1. e4 e5 2. Nf3 Nc6 3. Bc4 is a draw with correct play.
1. f3 e5 2. Kf2 is also a draw with correct play. What matters is when theory ends and a novelty is played forcing you to think, what side of the board would you rather play. The "true score" would be positive if you'd pick white, and negative if you'd pick black. There's an objective best choice, and the position being a tablebase draw (correct play?) is irrelevant, because we don't have 32men tbs.
Great spirits have always encountered violent opposition from mediocre minds.

Post Reply