matejst wrote: ↑Sat Jun 20, 2020 2:11 am
Brendan,
Tal looked great also against Smislov, Panno, Spassky, Fisher, Botvinnik...
Hey Boban, long time no talk, my friend...
Of course I know Tal also looked great against those guys...but I am talking about extremes.
Observe this game for example...
[pgn][Event "Simul"]
[Site "Chicago, IL USA"]
[Date "1988.??.??"]
[EventDate "?"]
[Round "?"]
[Result "1-0"]
[White "Mikhail Tal"]
[Black "Jack Miller"]
[ECO "C55"]
[WhiteElo "?"]
[BlackElo "?"]
[PlyCount "65"]
1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.d4 d6 5.dxe5 Nxe4 6.Bxf7+ Kxf7
7.Qd5+ Be6 8.Qxe4 Be7 9.O-O d5 10.Qd3 Qd7 11.Re1 Raf8 12.Nc3
Ke8 13.Ng5 Bc5 14.Nxe6 Bxf2+ 15.Kh1 Bxe1 16.Nxf8 Rxf8 17.Bg5
Nb4 18.Qe2 Nxc2 19.e6 Qd6 20.Nb5 Qe5 21.h4 Qg3 22.Rd1 Rf2
23.Qxf2 Bxf2 24.Rxd5 Qxh4+ 25.Bxh4 Bxh4 26.Nxc7+ Kf8 27.Rf5+
Bf6 28.Rd5 a5 29.Rd7 Nb4 30.Rf7+ Kg8 31.Rxf6 Nc6 32.Rf7 g6
33.e7 1-0[/pgn]
Tal did not play such sacrifice-filled, magician-like games against elite opponents.
Nice games, but not amazing like this...their defensive level is too high.
Same goes for engines.
matejst wrote: ↑Sat Jun 20, 2020 2:11 am
But back to the topic... I do not know if I would call it "style", but side-by-side analysis allows you to spot differences among engines. Then, watching matches between some weaker engines I like and the beast rated 3200+, games are decided most often at the end of the middlegame when an advantage in search depth of 20+ plies makes all the difference. In the opening, despite Ed (Shroeder)'s results, I find the new engines, using Texel tuning, in general weaker. After the moves 1. d4 Nf6 2. c4 e6 3. Nf3 Bb4+ 4. Bd2 Qe7 5. g3 Nc6 engines like Xiphos, Ethereal, will chose 6.e3 even at depths 20+, Wasp 4.0 too (when Wasp 1.02 would chose Sc3).
Texel tuning-like methods brought something new, but they have their limits, and, without improvement and evolution, I think they are detrimental for further improving engines. I would add --
most modern engines play a similar ugly chess, and it is especially noticeable in well analyzed quiet opening positions. Engine moves often makes no sense. One could say that engine moves make no sense to me -- but when I switch to SugaR NN, Winter, RubiChess (although Rubi was also trained with the TexTM; I would be very interested to hear from the author if he tuned his engine by hand too), suddenly they chose the moves from opening books, or moves with plans I understand from similar positions.
For me, it is clear now that the future is a mixture of nets and alpha/beta, with learning/training abilities, a right balance between search and evaluation. So far, the evaluation is lagging behind.
I agree for the most part here too. Modern engines play super strong, but boring and weird chess.
Although I do like messing around with a few of these Leela nets.
I also still prefer Wasp 1.02 as well.