The colder days : prime time for new engines

In the colder days we need new engines to warm us in front our pcs.

Engines that play different.
I am not interested in stockfish.

Thorsten, I often analyse with two engines at the same time and compare their lines. The differences are not big, especially in the openings. It is a bit more interesting when one uses multi-pv, or compares a slow engine with a very fast engine, or a NN-engine with a HCE-engine. It helps finding interesting plans in a given position, and understand it better.

In general, using the same techniques, the same tuning methods, similar data, engines are very similar. A few could evolve differently: Seer, Berserk are my first candidates. The first has a very interesting NN, the other a good combination of fast search and good HCE evaluation, soon to be combined with smaller nets for parts of the eval, like SlowChess. I liked Winter too, but the development has stalled lately.

All those engines are that similar.

I have completely stopped pc chess because no engine interests me anymore.
They are obviously state if the art, boring engines.

If i would not have the old dedicated and the old dos engines etc, i could stop doing computerchess.

It is all hardware hardware hardware.
Auto testing . Auto tuning. Its imo boring.
Thats not chess. Its increasing ELO. But these engines rarely play chess.

I know this sounds like a paradoxon.

They simply lack options. Some older projects had personalities -- one could immediately notice a difference in play, more positional or more aggressive plans and moves. I don't know about the new Komodos, the one with personalities are still commercial, but when I load Weiss, Berserk or Koivisto 4.83 (all three have HCE), I see the same lack of options (almost none; online book or online TBs), and I feel that the only differences are the way they prune and the depths they achieve.

I compared ShashChess Petrosian and ShashChess Tal in different positions: the differences are quite small. The emphasis is nowadays mostly on search, and chess engines are just an engineering feat nowadays. I used to test different versions of Komodo at fixed depths: Komodo 8, although tactically weaker, has about the same strength as K12 at depth 16. I feel that the evaluation function has been neglected. Even with the NNs, I think that this job has not been taken seriously. Fat Titz is a paradigmatic case of the new philosophy: "the biggest the net, the better the eval". It seems that a combination of nets and HCE could give better results and more original results, but nobody tries.

They concentrate on ELO.
And that makes it boring.