Hi... I'm working on engine research

[jkominek]

I'll be honest, phrases like "epistemic humility and flexibility" confuse me more than anything.

He is just pulling every empty comment through an obfuscating LLM...

LLM filtered or not, I have noticed "epistemic" has been trending lately in the vernacular. Such things used to be called ten dollar words.

epistemic humility = a fancy way of saying I claim not to know the answer (or trying not to act like a know-it-all jackass)
epistemic flexibility = a fancy way of saying I change my mind a lot

[chrisw]

Training a network for exactly reproducing win probability in principle desroys the possibility for the engine to convert a non-trivial win. E.g. a KBNK mate typically takes some 60 ply, but, as long as no B or N is blundered away, is always a certain win. There is no way a search would reach 60 ply if it has no guidance for what to prune; without pruning you must be happy to reach 8 ply. And there is no guidance, neither for prunig as for striving for intermediate goals (like getting the bare King into the corner), as all non-sacrificial end-leaves will have an identical 100% score. This reduces the engine to a random mover biased against giving away any material. No way its random walk through state space will ever bring a checkmate within the horizon. At some point it will stumble into the 50-move barrier, and there is nothing it can be done to avoid it.

A corrolary is that when an engine is trained to reproduce the true win rate on the games played by the latest version, it can never achieve the exact win rate. When it is still ignorant it will discover that it will have a higher chance of winning in positions where the mate is close, just because it has a larger chance to stumble on a position that has it within the search horizon. Those will then get higher score, which can be seen by the search from further away, so that the win rate there will also improve. But it has to keep a score gradient large enough to guide the play towards the mate from far away. And there is some minimum gradient for which this still works, because there will always be evaluation noise. So you will get some equilibrium, where positions far away from any mate only get, say, 90% win rate. Because that will indeed be their win rate, because the gradient from 90% to 100% is spread over so many moves that it has become so small that it fails to follow it in 10% of the cases.

Are you saying an engine can have precise, accurate analysis, or dominating play strength, but not really both at once?

The enemy's king is hiding in a desert with lots of high dunes obscuring vision at ground level. Two different companies of soldiers will attempt to kill him:

1. Stockfish (SF)

2. Quick checkmater (QC)

SF parachute into the middle of the desert, quickly getting close. Now they're there, though, they don't know which direction to travel in: they just have to guess, try a direction, and hope to uncover some clues.

QC start their assault from the edge of the desert, carefully charting their position as they go. This will lead them directly to the king - but it's too obvious and predictable, so they will get ambushed and killed.

Unbelievable BS. Chess is not life nor war, both of which are massively complex, unbounded and anything can happen. They are neither two sided, take it in turns to move operating on a tiny space of 64 squares with pieces that can move in six ways only. The "chess is life" concept is and always was a nonsense fantasy. Tal included.

Simple game theory concepts don't necessarily capture the complexity of chess. Chess may have nonlinear, emergent properties that only arise at different scales. Seeing chess as only 64 squares and 16 pieces is just one way to look at the resulting game.

There may be more emergent heuristics, which is more likely what you mean. Whether normal chess players can understand them though is another matter. Probably the best young talent have developed their own “heuristics” by watching/playing engines but again whether these are going to be generally understandable is questionable.

You have to remember it’s two alternating sides with 64 squares and six move types. Compare and contrast to life or war, the other favourite. No comparison. The examples given in prior posting are just nonsense.

[chrisw]

Training a network for exactly reproducing win probability in principle desroys the possibility for the engine to convert a non-trivial win. E.g. a KBNK mate typically takes some 60 ply, but, as long as no B or N is blundered away, is always a certain win. There is no way a search would reach 60 ply if it has no guidance for what to prune; without pruning you must be happy to reach 8 ply. And there is no guidance, neither for prunig as for striving for intermediate goals (like getting the bare King into the corner), as all non-sacrificial end-leaves will have an identical 100% score. This reduces the engine to a random mover biased against giving away any material. No way its random walk through state space will ever bring a checkmate within the horizon. At some point it will stumble into the 50-move barrier, and there is nothing it can be done to avoid it.

A corrolary is that when an engine is trained to reproduce the true win rate on the games played by the latest version, it can never achieve the exact win rate. When it is still ignorant it will discover that it will have a higher chance of winning in positions where the mate is close, just because it has a larger chance to stumble on a position that has it within the search horizon. Those will then get higher score, which can be seen by the search from further away, so that the win rate there will also improve. But it has to keep a score gradient large enough to guide the play towards the mate from far away. And there is some minimum gradient for which this still works, because there will always be evaluation noise. So you will get some equilibrium, where positions far away from any mate only get, say, 90% win rate. Because that will indeed be their win rate, because the gradient from 90% to 100% is spread over so many moves that it has become so small that it fails to follow it in 10% of the cases.

Are you saying an engine can have precise, accurate analysis, or dominating play strength, but not really both at once?

The enemy's king is hiding in a desert with lots of high dunes obscuring vision at ground level. Two different companies of soldiers will attempt to kill him:

1. Stockfish (SF)

2. Quick checkmater (QC)

SF parachute into the middle of the desert, quickly getting close. Now they're there, though, they don't know which direction to travel in: they just have to guess, try a direction, and hope to uncover some clues.

QC start their assault from the edge of the desert, carefully charting their position as they go. This will lead them directly to the king - but it's too obvious and predictable, so they will get ambushed and killed.

Unbelievable BS. Chess is not life nor war, [..] The "chess is life" concept is and always was a nonsense fantasy. Tal included.

Says the guy who once wrote:
(I do quote Mr. Chris W. In person:)

„ Von Manstein
============
If, as is said, chess is war, then there must be lessons to be learnt from military history. I have already alluded to the static, boring First World War style of the classical programs (and their programmers !). The opposite style can be found in several histories, Rommel in North Africa, Alexander the Great against Darius, Von Manstein in Russia. Alexander, despite being outnumbered many times, concentrated the powerful mobile part of his army, attacked the stronger Persians, cut through and went straight for Darius himself. The bulk of Darius's army was not engaged, but the battle was decisively won - a classic king attack. Von Manstein (and Rommel) both understood that the power of the outnumbered German army lay in superior staff work, concentration of forces, striking blows to knock the enemy off balance. The looking-glass chess program must contain knowledge of these dynamic elements; and it is only the looking-glass program that has the knowledge and evaluation time available to calculate such ephemerals.“

So suddenly this was all wrong ?

I can’t believe you argue against yourself.
What has happened to you ?!

This is a quote from THIS forum from 2007.

„ In a materialistic world, in the materialistic world of computer chess, Chess System Tal offers the alternative pathway of idealism.“

What’s the problem? In the environment back then most engines were not doing much more than adding up piece square tables. Obviously an HCE which knew some ways to measure multiple piece power being exerted against the opponent king despite being at an overall material disadvantage would find ways to sacrifice material in exchange for attacking opportunities and win games this way. Asymmetric strategy.
You can draw lessons downwards, from life or war or whatever down to chess, but you can’t draw them in the other direction. No number of chess heuristics are going to help you in life. Bishop pair has no analogous life situation, nor doubled pawns nor any other heuristic, complex or not.

[FireDragon761138]

Training a network for exactly reproducing win probability in principle desroys the possibility for the engine to convert a non-trivial win. E.g. a KBNK mate typically takes some 60 ply, but, as long as no B or N is blundered away, is always a certain win. There is no way a search would reach 60 ply if it has no guidance for what to prune; without pruning you must be happy to reach 8 ply. And there is no guidance, neither for prunig as for striving for intermediate goals (like getting the bare King into the corner), as all non-sacrificial end-leaves will have an identical 100% score. This reduces the engine to a random mover biased against giving away any material. No way its random walk through state space will ever bring a checkmate within the horizon. At some point it will stumble into the 50-move barrier, and there is nothing it can be done to avoid it.

A corrolary is that when an engine is trained to reproduce the true win rate on the games played by the latest version, it can never achieve the exact win rate. When it is still ignorant it will discover that it will have a higher chance of winning in positions where the mate is close, just because it has a larger chance to stumble on a position that has it within the search horizon. Those will then get higher score, which can be seen by the search from further away, so that the win rate there will also improve. But it has to keep a score gradient large enough to guide the play towards the mate from far away. And there is some minimum gradient for which this still works, because there will always be evaluation noise. So you will get some equilibrium, where positions far away from any mate only get, say, 90% win rate. Because that will indeed be their win rate, because the gradient from 90% to 100% is spread over so many moves that it has become so small that it fails to follow it in 10% of the cases.

Are you saying an engine can have precise, accurate analysis, or dominating play strength, but not really both at once?

The enemy's king is hiding in a desert with lots of high dunes obscuring vision at ground level. Two different companies of soldiers will attempt to kill him:

1. Stockfish (SF)

2. Quick checkmater (QC)

SF parachute into the middle of the desert, quickly getting close. Now they're there, though, they don't know which direction to travel in: they just have to guess, try a direction, and hope to uncover some clues.

QC start their assault from the edge of the desert, carefully charting their position as they go. This will lead them directly to the king - but it's too obvious and predictable, so they will get ambushed and killed.

Unbelievable BS. Chess is not life nor war, both of which are massively complex, unbounded and anything can happen. They are neither two sided, take it in turns to move operating on a tiny space of 64 squares with pieces that can move in six ways only. The "chess is life" concept is and always was a nonsense fantasy. Tal included.

Simple game theory concepts don't necessarily capture the complexity of chess. Chess may have nonlinear, emergent properties that only arise at different scales. Seeing chess as only 64 squares and 16 pieces is just one way to look at the resulting game.

There may be more emergent heuristics, which is more likely what you mean. Whether normal chess players can understand them though is another matter. Probably the best young talent have developed their own “heuristics” by watching/playing engines but again whether these are going to be generally understandable is questionable.

You have to remember it’s two alternating sides with 64 squares and six move types. Compare and contrast to life or war, the other favourite. No comparison. The examples given in prior posting are just nonsense.

Have you heard of Philidor, "pawns are the soul of chess"? A pawn chain forms something greater than the sum of its parts, due to the interrelationships between the pieces. Philidor understood that. That's something that a surface reading of game theory won't tell you.

[chrisw]

Training a network for exactly reproducing win probability in principle desroys the possibility for the engine to convert a non-trivial win. E.g. a KBNK mate typically takes some 60 ply, but, as long as no B or N is blundered away, is always a certain win. There is no way a search would reach 60 ply if it has no guidance for what to prune; without pruning you must be happy to reach 8 ply. And there is no guidance, neither for prunig as for striving for intermediate goals (like getting the bare King into the corner), as all non-sacrificial end-leaves will have an identical 100% score. This reduces the engine to a random mover biased against giving away any material. No way its random walk through state space will ever bring a checkmate within the horizon. At some point it will stumble into the 50-move barrier, and there is nothing it can be done to avoid it.

A corrolary is that when an engine is trained to reproduce the true win rate on the games played by the latest version, it can never achieve the exact win rate. When it is still ignorant it will discover that it will have a higher chance of winning in positions where the mate is close, just because it has a larger chance to stumble on a position that has it within the search horizon. Those will then get higher score, which can be seen by the search from further away, so that the win rate there will also improve. But it has to keep a score gradient large enough to guide the play towards the mate from far away. And there is some minimum gradient for which this still works, because there will always be evaluation noise. So you will get some equilibrium, where positions far away from any mate only get, say, 90% win rate. Because that will indeed be their win rate, because the gradient from 90% to 100% is spread over so many moves that it has become so small that it fails to follow it in 10% of the cases.

Are you saying an engine can have precise, accurate analysis, or dominating play strength, but not really both at once?

The enemy's king is hiding in a desert with lots of high dunes obscuring vision at ground level. Two different companies of soldiers will attempt to kill him:

1. Stockfish (SF)

2. Quick checkmater (QC)

SF parachute into the middle of the desert, quickly getting close. Now they're there, though, they don't know which direction to travel in: they just have to guess, try a direction, and hope to uncover some clues.

QC start their assault from the edge of the desert, carefully charting their position as they go. This will lead them directly to the king - but it's too obvious and predictable, so they will get ambushed and killed.

Unbelievable BS. Chess is not life nor war, both of which are massively complex, unbounded and anything can happen. They are neither two sided, take it in turns to move operating on a tiny space of 64 squares with pieces that can move in six ways only. The "chess is life" concept is and always was a nonsense fantasy. Tal included.

Simple game theory concepts don't necessarily capture the complexity of chess. Chess may have nonlinear, emergent properties that only arise at different scales. Seeing chess as only 64 squares and 16 pieces is just one way to look at the resulting game.

There may be more emergent heuristics, which is more likely what you mean. Whether normal chess players can understand them though is another matter. Probably the best young talent have developed their own “heuristics” by watching/playing engines but again whether these are going to be generally understandable is questionable.

You have to remember it’s two alternating sides with 64 squares and six move types. Compare and contrast to life or war, the other favourite. No comparison. The examples given in prior posting are just nonsense.

Have you heard of Philidor, "pawns are the soul of chess"? A pawn chain forms something greater than the sum of its parts, due to the interrelationships between the pieces. Philidor understood that. That's something that a surface reading of game theory won't tell you.

A few cursory thoughts will tell you that interaction and cooperation between the different piece types and with each other is consequential. Pawns are the soul of chess is a rambling bit of philosophy that sounds good as a book title but doesn’t do much else. I played e4 because “pawns are the soul of chess”? Haha. What’s your point, other than to obfuscate?

[hgm]

I guess there is a little bit more to that. The presence of a piece like the Pawn has a huge impact on the character of the game. The essential property is that it is a weak, vulnerable piece (often a 'sitting duck'), with an extremely powerful promotion. Chess variants that do not have that are very drawish. As we know from Pawnless end-games; these are much harder to win. Or from Shatranj, where the Pawns promote to a piece that is hardly better than a Pawn.

The typical dynamics of Chess is that positional advantages at some point accumulate to a value where you can gain a relatively defenseless Pawn. This then can result with only equal trading in positions where the strong player has a passer. The opponent than has to dedicate part of his (non-pawn) piece power to stopping that passer, giving him more tactical disadvantage versus the player that can use his pieces freely and let the tactically almost useless Pawns shield the opponent from promotion. This tactical advantage can then be converted to gaining more Pawns, etc.

This cycle completely breaks down when there are no promoting pieces, when they are too mobile to easily gain them, when they are too mobile to stop each other from promotion, etc. So yes, the presence of a piece with exactly the move and promotion of the FIDE Pawn to a very large extent determins the dynamics of a typical chess game. Irrespective of how exactly the other pieces move. It doesn't matter much whether the strongest piece moves like Q = R + B, or like R + N. Or whether you have a Knight or a piece that steps one orthogonally and jumps two diagonally. The games will be different, for sure, but the general strategy stays the same. But if you replace Pawns by Scorpions (which in addition have the two widest forwar Knight moves), or Berolinas (which move diagonally and capture straight ahead, so that they hardly ever block each other), you get completely different game dynamics.

[FireDragon761138]

We have finished making our project, Theoria-SF with the Stockfish 16.1 engine, now we are attempting to see what we can do with Stockfish 17.1. Whichever one has a higher play strength, we will probably end up distributing. Stockfish 17.1 is harder to work with as it requires more GPU memory, and I'm not sure that threat input will add measurably to evaluation strength of the engine.

[towforce]

We have finished making our project, Theoria-SF with the Stockfish 16.1 engine, now we are attempting to see what we can do with Stockfish 17.1. Whichever one has a higher play strength, we will probably end up distributing. Stockfish 17.1 is harder to work with as it requires more GPU memory, and I'm not sure that threat input will add measurably to evaluation strength of the engine.

What changes have you made to Stockfish 16.1?

[chrisw]

We have finished making our project, Theoria-SF with the Stockfish 16.1 engine, now we are attempting to see what we can do with Stockfish 17.1. Whichever one has a higher play strength, we will probably end up distributing. Stockfish 17.1 is harder to work with as it requires more GPU memory, and I'm not sure that threat input will add measurably to evaluation strength of the engine.

Usually they change the name id


What changes have you made to Stockfish 16.1?

[FireDragon761138]

We have finished making our project, Theoria-SF with the Stockfish 16.1 engine, now we are attempting to see what we can do with Stockfish 17.1. Whichever one has a higher play strength, we will probably end up distributing. Stockfish 17.1 is harder to work with as it requires more GPU memory, and I'm not sure that threat input will add measurably to evaluation strength of the engine.

What changes have you made to Stockfish 16.1?

Different NNUE training, also using only a single NNUE instead of two.

I've thought about disabling optimism... in future versions I might make it a UCI option. But I think probably on the whole its beneficial and doesn't detract from the design goals. As near as I can tell, optimism in Stockfish's engine emulates the inherent optimism in MCTS search, making it less "minimax" like.