I am really looking forward to the day when the full 8-piece tablebase is officially finished. That will be the next big step toward getting closer to solving chess. Thanks a lot for the update and the blog post.
I am really looking forward to the day when the full 8-piece tablebase is officially finished. That will be the next big step toward getting closer to solving chess. Thanks a lot for the update and the blog post.
Waiting for someone to download these with SSD and RAM prices skyrocketing... :)
Waiting for someone to download these with SSD and RAM prices skyrocketing... :)
@tpr said:
Very interesting, a leap forward.
8 men is more relevant than 7 men, as 8 men can be materially balanced and 7 men not.
for building positional differential intuition? (find the little positional change that makes the diff).
I might be silly, but that is often my questionning, instead of seeking the huge win... I like to do mini-games like that, of course there is the full game from standard initial position... but within our small brain ability to encompass sets of critters on the board..
like zugzwang strict definition requires us to see all the outcomes to agree to its status.. otherwise it is just an impression... relative to what we can foresee.
@tpr said:
> Very interesting, a leap forward.
> 8 men is more relevant than 7 men, as 8 men can be materially balanced and 7 men not.
for building positional differential intuition? (find the little positional change that makes the diff).
I might be silly, but that is often my questionning, instead of seeking the huge win... I like to do mini-games like that, of course there is the full game from standard initial position... but within our small brain ability to encompass sets of critters on the board..
like zugzwang strict definition requires us to see all the outcomes to agree to its status.. otherwise it is just an impression... relative to what we can foresee.
Even in 2025, sharing large amounts of data on a budget is quite challenging, especially when neither Marc in the US nor Niklas in Europe had fiber internet in their homes. We considered various providers who would pick up data to upload, or would allow bringing data on-site to upload, but ultimately settled on sending three hard drives in a box.
Reminded me of this famous quote from my networking course -
"Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway." - Andrew S. Tanenbaum
> Even in 2025, sharing large amounts of data on a budget is quite challenging, especially when neither Marc in the US nor Niklas in Europe had fiber internet in their homes. We considered various providers who would pick up data to upload, or would allow bringing data on-site to upload, but ultimately settled on sending three hard drives in a box.
Reminded me of this famous quote from my networking course -
"Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway." - Andrew S. Tanenbaum
This is awesome progress to see. I'm wondering if this has been battle tested in a computer arena? E.g. Stockfish without OP1 vs Stockfish with it? What's the Elo gain? I understand tablebases are interesting for many reasons aside form engine performance, but was curious about this aspect! Another way to think about it - does the existence of this tablebase change the top move in Stockfish for preceding positions? What can we learn from those positions?
This is awesome progress to see. I'm wondering if this has been battle tested in a computer arena? E.g. Stockfish without OP1 vs Stockfish with it? What's the Elo gain? I understand tablebases are interesting for many reasons aside form engine performance, but was curious about this aspect! Another way to think about it - does the existence of this tablebase change the top move in Stockfish for preceding positions? What can we learn from those positions?
I don't believe Stockfish supports probing these new tablebases yet , but even if it did (I guess it will at some point), it's likely not worth much elo for Stockfish. 7-men is only worth a few elo points, and you need everything to be on SSD for the probing speed to be acceptable. If you put the tablebases on a hard drive, you actually lose elo because the search is slowed down.
Another side effect of tablebases is that, in drawn positions, it can push the engine away from the most challenging lines because it starts seeing draws everywhere and stop being able to differentiate easy from hard to hold positions.
I don't believe Stockfish supports probing these new tablebases yet , but even if it did (I guess it will at some point), it's likely not worth much elo for Stockfish. 7-men is only worth a few elo points, and you need everything to be on SSD for the probing speed to be acceptable. If you put the tablebases on a hard drive, you actually lose elo because the search is slowed down.
Another side effect of tablebases is that, in drawn positions, it can push the engine away from the most challenging lines because it starts seeing draws everywhere and stop being able to differentiate easy from hard to hold positions.