If you are a correspondence player, you need Leela now! - A corr GM opinion on Leela

drmrboss

null

LouStule
What’s it say? The print is too small. Read it to me. Thanks.
nighteyes1234
LouStule wrote:
What’s it say? The print is too small. Read it to me. Thanks.

 

It says that Tansel Turgut thinks Leela is better than Stockfish.

He gives no examples. He gives no other proof. Trust him.

P.S. Just to show you how much this guy knows, he thinks best would be a NN + AB engine...in other words he wants to mix Leela with SF to get the best, even though SF is inferior.

LouStule
Thanks but why is he telling this to Daily players? Using an engine is against the rules.
drmrboss

Stockfish developers are really struggling to improve SF in blocked positions. (e.g French Advanced, SF is really dying against Leela attack)

 

However SF is a boss in tactical positions, open positions and endgame with Tablebase assess.

 (copy from a Stockfish developer)

 

I have noticed that in closed center games like advanced french, Kings indian etc. SF and AB engines in general are vulnerable. This is due to the strategic nature of the positions, which requires maneuvering, long term planning, pawn breaks and positional slow motion attacks. Everything that our evaluation tells us its good, and we go for it, could be a bad strategic decision. ie. we could gain space on the Qside and get hefty bonuses, while our King sitting "safe", threatless on the other side is trapped like a mouse in the corner, with a karma of a certain death, but too distant to take notice. MCTS is strong at this due to simulation of playouts to the end, most of Leela wins vs SF are scored in such positions.

So what can we do to fortify this inherently weak area of the AB, positions so deep that search does not suffice?

Imo by changing evaluation terms and parameters we will have limited success, we might become better at those positions and worse at others.

My idea is to first define the subset of the particular positions.
 
A definition could just be: D and E files both having locked pawns (W d4 e5 - B d5 e6 etc) and afterwards tune eval and search specifically for this subset.

I imagine that a deeper tree search is required (thus less wide), less pruning on pawn moves as they are crucial (especially pawn breaks: on files next to the locked ones (ie C, F)), mobility losing importance, K getting a great bonus for being on the side of space advantage (or a penalty for being on the advanced enemy side). In fact I consider the categorization of space in sides a crucial element for the outcome: The space advantage on the side of the enemy K to be much more valuable. But we somehow have to also ensure that this side is not blocked dead.

Those ideas could also be tried generically, but imo there is great potential into specialized treatment.

In summery, reaching extreme depth is the key, piece play (shuffling) is interchangeable thus a wasteful allocation of resources preventing the solution of the position through pawn moves, King is the bounty.

 

IMBacon

I do not ,and will not use an engine for correspondence chess even though it is allowed.

Johnny_Gumshoe
I did not know it was allowed and quite frankly alarmed that it is.
IMBacon
Johnny_Gumshoe wrote:
I did not know it was allowed and quite frankly alarmed that it is.

This is why i will not join the International Correspondence Chess Federation.  I have no desire to know whos engine is better.  I want to play someone one on one.  Just us humans....

LouStule
Agreed. What’s the point of playing if everyone is using an engine? Bragging rights I guess “My computer is smarter than your computer”.
stiggling
LouStule wrote:
Agreed. What’s the point of playing if everyone is using an engine? Bragging rights I guess “My computer is smarter than your computer”.

Because people who just copy an engine don't do very well, otherwise the competition wouldn't exist at all.

Daniel1115

LouStule wrote:

Agreed. What’s the point of playing if everyone is using an engine? Bragging rights I guess “My computer is smarter than your computer”.

My computer has more Ram than yours. Alternatively:I have no life and can run my engine the whole day every day(or whatever the interval is)

Taskinen

The problem is that it would be way too difficult for serious correspondence chess to try to prevent engine use. Every case of someone possibly cheating with an engine would be extremely hard to prove, and it could easily backfire on people playing good games without an engine to be claimed as cheaters. That's the main reason most of the serious live chess events are played on a real board, and it's usually only speed chess that is played online. In speed chess it's practically impossible to play engine like accuracy, compared to how top GM's play classical live chess (and correspondence games are obviously even more accurate). And since you were allowed to use books and all other stuff with correspondence games anyways, it's really difficult to measure after the game which moves were generated by engine, and which were results of careful study, tablebases, books or whatever. So frankly, correspondence chess is just playing a game of chess where you are allowed to use anything to find the moves. And that is the only way they can keep it in the era where engines are stronger than humans. I know it's a disappointing direction for many who enjoyed playing correspondence chess pre engine era purely human vs. human.

Taskinen
Johnny_Gumshoe wrote:
I did not know it was allowed and quite frankly alarmed that it is.


In chess.com daily games using engine to evaluate moves is not allowed. But in serious rated correspondence games, engine use is allowed.

Pulpofeira

I don't think it is only about who has the better engine anyway, some people are best at taking the most out of them.

IMBacon
Kilbroney wrote:
IMBacon wrote:
Johnny_Gumshoe wrote:
I did not know it was allowed and quite frankly alarmed that it is.

This is why i will not join the International Correspondence Chess Federation.  I have no desire to know whos engine is better.  I want to play someone one on one.  Just us humans....

Do I detect an overt snobbery or even a self assigned purist elitism from some users on this august server against those who have joined ICCF ? I know the ICCF is recognised by FIDE and you can real meaningful rating points from the comfort of your own home. Just wonder how many of the account holders here have dual membership ? 

No snobbery at all.  If i want to play an engine, i can do that on my laptop.  I prefer playing people.

pfren

Good marketing tricks by Nvidia. I applause.

 

aydenmerritt

You play 

 

Chessbabe-youtube
MyGreatMethod1 wrote:

Leela is too weak all sales talk. Not advisable for analysis either. Yes Leela may be as strong as Magnus Carlsen but too weak for Stockfish and Komodo.

Currently #3 in CCCC2 with just a point behind Stockfish. Defeated Komodo in a 30 game rapid match to secure 3rd place in CCCC1. Are you living under a rock or something? Leela is already as good as Komodo and Houdini and closing in fast on Stockfish.

https://www.chess.com/computer-chess-championship

d0gofch3ss
Chessbabe-youtube wrote:
MyGreatMethod1 wrote:

Leela is too weak all sales talk. Not advisable for analysis either. Yes Leela may be as strong as Magnus Carlsen but too weak for Stockfish and Komodo.

Currently #3 in CCCC2 with just a point behind Stockfish. Defeated Komodo in a 30 game rapid match to secure 3rd place in CCCC1. Are you living under a rock or something? Leela is already as good as Komodo and Houdini and closing in fast on Stockfish.

https://www.chess.com/computer-chess-championship

Leela is good, but still not yet 'better than houdini'. With your logic, houdini did better than leela in cccc1, and is leading ahead of houdini in ccc2. In playing strength, she will eventually surpass sf, but still lacks hugely in analysis. 

But you are, for the most part, correct. Can't trust someone to be an engine expert when they're comparing them to Carlsen happy.png

DamonevicSmithlov

This morning I watched a good video on artificial intelligence and of course it talked about Alpha Go and Alpha Go Zero and the long term implications of such technology. The video even showed Kasparov walking away from the board in disgust after a game with Deep Blue. The videos main idea was that it wanted us to think of the broader picture when implementing this technology and the fascinating and/or frightening possibilities. Elon Musk says at the start of the video: "It could be terrible and it could be great (referring to the future of AI). It's not clear. But one thing is for sure, we will not control it."