Stockfish and Multi-PV

[Rebel]

Lots of truth in there. We need a CCRL shorty rating list. A couple of years ago I (out of curiosity) played a shorty match between 2 equal rated versions of Stockfish and Komodo and the latter was crushed. And I concluded Stockfish had the better opening code. Now I know better lol. What you said, the last bastion. We need a new competition to tickle the programmers.

[MikeB]

Lots of truth in there. We need a CCRL shorty rating list. A couple of years ago I (out of curiosity) played a shorty match between 2 equal rated versions of Stockfish and Komodo and the latter was crushed. And I concluded Stockfish had the better opening code. Now I know better lol. What you said, the last bastion. We need a new competition to tickle the programmers.

Stockfish could use an opening phase function in its evaluation. My thought was that HIARCS always seemed to play pretty well without an opening book - but that was from observations years ago - not sure today If i would have the same opinion.

[Rebel]

One idea would be to create a STS type of opening set rating engines with MEA from Ferdy.

Such as:
1.d4 and 1.e4 10 points
1.c4 and 1.Nf3 8 points
1.f4 6 points

etc.

http://rebel13.nl/rebel13/experiments.html#six

[jp]

I think I am wasting my time looking at the moves it sometimes produces.

This situation happened because engine testers chose to play with generic books that take over the opening and play reasonable moves for the engine, and when the engine has to play it already skipped the problem entirely.

Engine developers have had no reason whatsoever to get better at the opening, and self-play has had a similar resolution. The elo chase is real

InfinityChess's Chess Engine Master tournament has shown that Leela isn't good at the openings either

That's why it's bad to give Leela tablebases. There needs to be incentive to improve its endgame play.

[Ovyron]

Such as:
1.d4 and 1.e4 10 points
1.c4 and 1.Nf3 8 points
1.f4 6 points

Hmmm, what recent developments in opening theory have showed me is that I'm not sure anymore if our preconceptions about what opening moves are best is good.

This started because of the bin book I downloaded that was bugged, and played 1.Nc3 no matter what, and I had a great performance with it. Then someone beat me with the black pieces, and I took a look, and noticed 1.Nc3 had 0% probability. I "fixed" it, and the book was now playing 1.e4 and 1.d4 as it should. And with those moves people keep beating me with the black pieces and have had the worst performance since then.

What if 1.Nc3 is a great move but people haven't found about it because it's an undiscovered gem? I have been using it to beat +2000 rated players on lichess.com and got 2nd place on < chess.com tourney, no way I could have done that with 1.e4 or 1.d4.

Should we give 1.Nc3 11 points in the test your propose? What if Rybka 2.2n2 (that loves the move) is best on the opening, but plays really badly the early middle game that nobody knows?

Maybe what we need is software like Aiquiry, that allows one engine to be used "as book", to play without book assistance until some point in the game, then switches to Stockfish, or some other at some point in the game, and we see how this new chess entity plays, maybe it'll play better than Stockfish without book all game, maybe some engine can push Leela's results beyond what we've seen. Maybe some engine without book with very low rating helps discover a new variation that is great that nobody has paid attention to because that engine misplays the positions.

The sky is the limit, but we have no idea of what openings are best, just what current entities are best at, but top Correspondence chess players, those that play at 2 years per game time control, have a preferrence of 1.c4 over 1.d4 or 1.e4, because at that level it's easier to with with c4. It's a bit mysterious.