towforce wrote: ↑Tue Aug 02, 2022 10:06 am
Where's the option for "SF will be the last engine to ever be the strongest one"?
IMO we will reach the upper limit of the elo scale before a stronger chess engine comes along.
^^ This.
The Elo is Just a Number created by us to differentiate the strength of the Engines, but the Elo scale could have infinity number.................................................................................................................................................?
Not possible: chess is a finite game.
But it will NOT be solved in your lifetime like Checker was, there are too many possible moves and combination that will require at least computer 50 times faster than Chinook, and endgame of at least 12 Pieces EGTB.
Chess won't be solved in the way that checkers was. However, there are other possibilities:
* stacking proofs on top of each other to prove the win of material is not possible from the opening position (and hence that chess is a draw, because the king is a subset of "material"). Software to prove things is steadily improving, or a chess specific set of proofs could be built
* discover key relationships between pieces that humans haven't found yet which always hold, and which enable any position to be evaluated accurately (NNs have already found valuable relationships which writers of hand-coded evaluation functions don't know - but I think we'll need a different technology to find the best relationships)
* even if chess isn't fully resolved, chess engines will be able to reach a level where they get beaten so rarely that no other engine can get 1 elo point higher (this is very likely to happen before chess gets formally resolved). At this point, we will be at the upper limit of the elo scale.
Writing is the antidote to confusion.
It's not "how smart you are", it's "how are you smart".
Your brain doesn't work the way you want, so train it!
CornfedForever wrote: ↑Tue Aug 02, 2022 7:45 pm
...and Dragon ( updates ) only ever seems to get close to the 'official' version of SF these days so it's often unfair to even compare a Dragon update with the current 'official version' of SF.
Without the 60 billion positions of leela data, our progress is a bit slower than SF’s. Still, slow but steady.
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".
dkappe wrote: ↑Thu Aug 04, 2022 5:33 am
Without the 60 billion positions of leela data, our progress is a bit slower than SF’s. Still, slow but steady.
I'm trying hard to produce self-play games...enough to obtain a reasonable strength...
running 2 systems almost 24/7.
I currently have 2.2 million games, about 220 million positions, w/ more being played every day and the nnue is currently at 2800 Elo.
dkappe wrote: ↑Thu Aug 04, 2022 5:33 am
Without the 60 billion positions of leela data, our progress is a bit slower than SF’s. Still, slow but steady.
You misread my post. Dragon does not and never has used the leela data. Our data all originated from Komodo 14 and Komodo 14.1 data and improved through reenforcement learning thereafter.
As for who makes the rules? As far as I know, only TCEC has some rather peculiar and inconsistent rules about engines, data and nets. Not sure what they would do if another engine other than Stockfish were to help themselves to the leela data.
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".
