Which of the many chess engines in this forum use b strategy ?

[mclane]

The neural networks play a better chess. Over the years watching machine chess on my arena GUI was pretty senseless because the games were, with exceptions, very boring.

But with LC0 this has changed.

But if we only learn out of it that we had to integrate NN into Stockfish to beat LC0 it is IMO a wrong essence.


The essence should be that we have to learn why LC0 plays that magic chess against the AB engines, even when those AB
engines were the champs of yesterday.

The quality level is higher.
It’s really better chess.

Now we have to observe and name it.
By examples.

[Dann Corbit]

I think that this is the golden moment we have been missing, and somehow now we are dismissing.
The one thing I have always wanted from chess engines is to teach me to play better chess.
I really don't care if engine X has a rating of 3875 and engine Y has a rating of 3893.
That means very, very little to me.
While I do learn a little from the constant beatings they give me out behind the wood shed, I wish there were a better way.
Maybe, the Neural networks have discovered some fundamentally new way to consider the chess board.
Maybe it is something that we could understand.
Maybe after we understood it, we could play better chess.
Maybe after we understood it, we could write better chess programs.
Maybe after we understood it, we could write better chess books.
Maybe after we understood it, we could watch a great chess match and understand better what was happening.

Maybe not, especially in my case, but I would like to know.

This is a fundamental flaw with NNs: there is no easy way to discern what factors are most important for a preference in a particular situation. If you wish to build a system that can explain what the most important factors are in a particular situation, you'd probably be best starting with a different technology.

Perhaps it is possible to hide chess facts from them when we learn and then deduce how they are using the information.
Or, perhaps there is a way to tag the neurons with what they are processing. A new sort of network.

[Dann Corbit]

...
We are at a crossroads where we are no longer teaching the engines how to play chess. In fact, they know much better that we do. It is time to start asking them questions so that we can understand what it is that we are missing.

Sure, unsubstantiated. Pure rumor and speculation. Perhaps pure, loggerheaded, leaden stupidity. But it makes sense to me (FWIW).

Good point, now we have "black-box" NNs, when do we start to un-black-box them?

--
Srdja

I think that this is the golden moment we have been missing, and somehow now we are dismissing.
The one thing I have always wanted from chess engines is to teach me to play better chess.
I really don't care if engine X has a rating of 3875 and engine Y has a rating of 3893.
That means very, very little to me.
While I do learn a little from the constant beatings they give me out behind the wood shed, I wish there were a better way.
Maybe, the Neural networks have discovered some fundamentally new way to consider the chess board.
Maybe it is something that we could understand.
Maybe after we understood it, we could play better chess.
Maybe after we understood it, we could write better chess programs.
Maybe after we understood it, we could write better chess books.
Maybe after we understood it, we could watch a great chess match and understand better what was happening.

Maybe not, especially in my case, but I would like to know.

You know. The best chess engine is 32 men EGTB. All you need is python-chess for movegen. A little bit of heuristic to randomly pick a move that wins/draws/loses. That's like 5 lines of compact Python code for perfect chess engine. That "engine" is really dumb but plays perfect chess. Chess isn't some AI research lab.

In Suicide chess white wins after 1. e3. It's all about remembering PV.

Yes, sure. And a 10 line minimax program can solve chess, given enough time. But of course neither of these proposals addresses anything that I have said.

[towforce]

I think that this is the golden moment we have been missing, and somehow now we are dismissing.
The one thing I have always wanted from chess engines is to teach me to play better chess.
I really don't care if engine X has a rating of 3875 and engine Y has a rating of 3893.
That means very, very little to me.
While I do learn a little from the constant beatings they give me out behind the wood shed, I wish there were a better way.
Maybe, the Neural networks have discovered some fundamentally new way to consider the chess board.
Maybe it is something that we could understand.
Maybe after we understood it, we could play better chess.
Maybe after we understood it, we could write better chess programs.
Maybe after we understood it, we could write better chess books.
Maybe after we understood it, we could watch a great chess match and understand better what was happening.

Maybe not, especially in my case, but I would like to know.

This is a fundamental flaw with NNs: there is no easy way to discern what factors are most important for a preference in a particular situation. If you wish to build a system that can explain what the most important factors are in a particular situation, you'd probably be best starting with a different technology.

Perhaps it is possible to hide chess facts from them when we learn and then deduce how they are using the information.
Or, perhaps there is a way to tag the neurons with what they are processing. A new sort of network.

We had a good discussion about this in the "solving chess" thread. I think that the evidence suggests that chess has an "emergent pattern" which would enable an evaluation function that could give a correct answer (win/draw/lose) on a static evaluation to be much smaller than most people would expect, and that it would be possible to find this emergent pattern.

However, despite the evidence, and a maddening feeling that this is very likely to be correct, it could be wrong, and it could be that the smallest evaluation function that would give a correct answer in most positions would actually have to be very large.

[mclane]

I think you have to be blind if you do not see how LC0 plays on a different planet then any AB program.
To be honest this is exactly the chess I had in mind when I spoke or helped Chris w. To do chess system tal.

An engine that find complex transactions between space, material positional aspects and tempi.

LC0 is doing these interchanges .

Only problem we can watch the games and be excited, but nobody really knows why it works or why the AB engines are so
Stupid to run in the Open Blade/fork and die.

But this is the chess we imaginated.

It’s a Paradoxon that this is been done by the machine itself. Where on the other side you had 40-50 or even longer history of mankind computerchess that leads to very strong engines like komodo or Stockfish, Houdini or rybka as the top head of the ELO rating lists.

The machines challenge the programmers in a field the human programmers were busy I a life time.

When should we expect a major breakthrough in science ?
When will a lone developer 'step through the looking-glass' ?
Who will this developer be ?

The machines (LC0) stepped through the looking glass and

Search - the lazy programmer's way to avoid evaluating a position.
==================================================================
The new paradigm differs from the classical by one simple conceptual switch.
The classical paradigm makes fast and simple evaluation at each node and generates intelligence from the search tree. The classical programmer looks for ways to make his search more efficient and his evaluation function simpler and faster. The 'looking-glass' paradigm makes slow and complex evaluations at each node and prefers to prune the search tree by use of this evaluation function. In this model search is to be avoided

LC0 has low NPS and the Clou of it is it’s evaluation that is folded into the huge NN.

Somehow it’s a rebuild of human chess knowledge normally folded into the brain in form of experience and chess-words/vocabulary, intuition.

B-Search or A-B-Search? - NO! Evaluation based or search based!
===============================================================
The classicists maintain the computer chess dichotomy of B-search (which I understand means pruning occurs at all levels of the tree) or A-B Search (which apparently means that part of the search is full width).
The looking-glass programmer condemns this dichotomy as meaningless.
The new paradigm makes the issue clear: chess programs either have simple evaluation and generate intelligence through search, or have complex evaluations and use limited search as a backup to cover oversights and mistakes. All chess programs prune in one way or another, but looking-glass programs, with complex evaluation, are able to prune more.

LCO is the new paradigm.
And the AB programs have to react.

In the Moment they react by implementing their own tiny nets into their experienced search.

[syzygy]

I am not a chess programmer. I mainly worked with others testing engines.
And there it was my job to find bugs or criticize things.

Without critical comments no progress. I don't see my comments as personal insults.

I am interested in computer chess as a whole thing. I do speak in general context, or do you see me criticize people ?

Yes, I do. You are being dismissive of the efforts of countless programmers. And on top of that you have no idea what you are talking about because all these programs are "Type B" and you absolutely do not have the faintest idea yourself how one should start creating something more "humanlike". I see nothing constructive in this long-lasting series of complaints fo yours (not just this thread), I see only whining.

The hardware progress often hides the fact that in this area almost nothing has happened.

Many hundreds of Elo gain by software alone is nothing? Do you consider yourself to have a clue?

[syzygy]

But if we only learn out of it that we had to integrate NN into Stockfish to beat LC0 it is IMO a wrong essence.

Why do you use the word "we" if you are obviously not a chess engine developer.

[mclane]

I am not a chess programmer. I mainly worked with others testing engines.
And there it was my job to find bugs or criticize things.

Without critical comments no progress. I don't see my comments as personal insults.

I am interested in computer chess as a whole thing. I do speak in general context, or do you see me criticize people ?

Yes, I do. You are being dismissive of the efforts of countless programmers. And on top of that you have no idea what you are talking about because all these programs are "Type B" and you absolutely do not have the faintest idea yourself how one should start creating something more "humanlike". I see nothing constructive in this long-lasting series of complaints fo yours (not just this thread), I see only whining.

The hardware progress often hides the fact that in this area almost nothing has happened.

Many hundreds of Elo gain by software alone is nothing? Do you consider yourself to have a clue?

If they would be b strategy they would sometimes give away a game with one big mistake.

How often is this with Stockfish ?

It’s not happening.
Stockfish uses a flexible rope when climbing that mountain, so it cannot fall that easily into the abyss.

B strategy programs use no rope.

The software made progress. But all this progress was not really used to increase the quality of chess. It increased the ELO.

Sorry if you feel attacked. This was not my intention.

[mclane]

But if we only learn out of it that we had to integrate NN into Stockfish to beat LC0 it is IMO a wrong essence.

Why do you use the word "we" if you are obviously not a chess engine developer.

Ah, come on, you can do better then this .


There was btw. an engine before that was capable to transform material and tactics and positional (space, mobility) into something, that was The King by Johan de koning.

[Tony P.]

After 142 moves of shuffling, Leela decided to go all-in and grab the pawn on b4. Then SF walking with a flexible rope was the one to sac the queen.

When I wish to run a home chess show, I'll just make a CPU Leela net play bullet vs some NNUE engine of similar strength and watch her blunder 10-move tactics.