4 ms·

This paper is pretty awful. Two amateur chess players guessed the best sequences of WBBWW and then plugged the result of their guesses into an engine to static

This paper is pretty awful. Two amateur chess players guessed the best sequences of WBBWW and then plugged the result of their guesses into an engine to statically analyse the position. There is absolutely no chance that they found the best options for either player. The sequencing argument that they make has no real foundation. Introducing double moves could have all kinds of side-effects which they later mention when showing their guesswork-variations but skip over when trying to make the previous "logical" argument. The only semi-useful sentence in this paper is the following: > More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White. The rest can be summarised as "We thought of this potential rule change and it seemed to us to be more even for black." If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair.

ummonk 5y ago
> If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair. This is what I was expecting them to do from seeing the abstract. Instead they just examined a handful of novel lines and evaluated them with Stockfish. An interesting proposed rules change, but a rather low quality analysis of it.
shmageggy 5y ago
Lichess has an API endpoint for their cached analysis, so they could have easily automated at least some of the new tree. Or they could have inserted a couple of `if` statements into Stockfish, recompiled, and simply analyzed the root position. Even calling this a "paper" is a stretch. I'm usually against twitter post threads as content but that seems about the level of analysis that's actually present here.
dmurray 5y ago
> More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White Also, there has been plenty of exploration around this in the recent past. "No-castling chess" was briefly popular in 2020 following some investigations by DeepMind and Vladimir Kramnik that found that training AlphaZero from scratch to play this led to more exciting games. [0] They didn't try this specific rule, or didn't publicize their results if they did. It's too different from chess to be of intrinsic interest to serious chess players. Maybe it's interesting as a variant in its own right! But this is the worst imaginable investigation of whether that's true. [0] https://www.chess.com/article/view/no-castling-chess-kramnik-alphazero https://www.chess.com/article/view/no-castling-chess-kramnik...
- Bootvis 5y ago
  The experiment is still ongoing. In July Anand and Kramnik (former WCs) played a no castling match: https://www.chess.com/news/view/anand-beats-kramnik-in-no-castling-match https://www.chess.com/news/view/anand-beats-kramnik-in-no-ca...
hakuseki 5y ago
> there are only around 10,000,000 possible game-states after the first WBBWW You don't even need to go that far. After WBBW you are already reverting to normal alternating play. Maybe that's around 400,000 positions?
beardyw 5y ago
> there are only around 10,000,000 possible game-states after the first WBBWW Do you have a source for that?

ummonk 5y ago

> If they had wanted to then there are only around 10,000,000 possible game-states after the first WBBWW and potentially looking at computer evaluations of the whole tree of possibilities could be semi-interesting and definitely very achievable. Probably better and easier to guess instead though to be fair. This is what I was expecting them to do from seeing the abstract. Instead they just examined a handful of novel lines and evaluated them with Stockfish. An interesting proposed rules change, but a rather low quality analysis of it.

shmageggy 5y ago

Lichess has an API endpoint for their cached analysis, so they could have easily automated at least some of the new tree. Or they could have inserted a couple of `if` statements into Stockfish, recompiled, and simply analyzed the root position. Even calling this a "paper" is a stretch. I'm usually against twitter post threads as content but that seems about the level of analysis that's actually present here.

dmurray 5y ago

> More light would be shed on this question if the two leading machine-learning chess programs, AlphaZero and Leela Chess Zero, were taught to play with our proposed change in the order of the 3rd and 4th moves from White-Black to Black-White Also, there has been plenty of exploration around this in the recent past. "No-castling chess" was briefly popular in 2020 following some investigations by DeepMind and Vladimir Kramnik that found that training AlphaZero from scratch to play this led to more exciting games. [0] They didn't try this specific rule, or didn't publicize their results if they did. It's too different from chess to be of intrinsic interest to serious chess players. Maybe it's interesting as a variant in its own right! But this is the worst imaginable investigation of whether that's true. [0] https://www.chess.com/article/view/no-castling-chess-kramnik-alphazero https://www.chess.com/article/view/no-castling-chess-kramnik...

Bootvis 5y ago
The experiment is still ongoing. In July Anand and Kramnik (former WCs) played a no castling match: https://www.chess.com/news/view/anand-beats-kramnik-in-no-castling-match https://www.chess.com/news/view/anand-beats-kramnik-in-no-ca...

hakuseki 5y ago

> there are only around 10,000,000 possible game-states after the first WBBWW You don't even need to go that far. After WBBW you are already reverting to normal alternating play. Maybe that's around 400,000 positions?

beardyw 5y ago

> there are only around 10,000,000 possible game-states after the first WBBWW Do you have a source for that?