It's a pity that it uses so little time however the guy who runs it said he considers writing a slow mode. Where the engine can think a lot longer. (though the user might have to wait a certain time between moves)
Right now it still blunders pieces left and right.
I think with a bit more strength, LCZero may become quite popular for people to play against, precisely because it doesn't play like a normal engine.
Judging from the experiences with Go, NN-based engines get strong at positional judgement much earlier than at tactics, which is kind of the opposite of Alpha-Beta engines (who can be tactical beasts but strategically clueless)
I tried it and I lost, which admittedly says more about my quality as a chess player than about Lczero's strength
Just played a game and won easily. I didn't realize I was a 3500 elo player
At first it was playing somewhat reasonably, giving me the impression of an almost computerish (albeit feable) opposition. But with 19... Qc8 ?? we see that tactics are not working well for lczero (yet!)
[pgn]
[White "lucas"]
[Black "lczero"]
1. d4 h5
2. e4 b6
3. Nf3 Bb7
4. d5 e6
5. c4 Nf6
6. Nc3 Qc8
7. Be2 Na6
8. a3 Nc5
9. Qc2 a5
10. Bg5 d6
11. O-O Be7
12. b4 axb4
13. axb4 Rxa1
14. Rxa1 Ncd7
15. Nb5 O-O
16. Nfd4 e5
17. Nf5 Bd8
18. Na7 Qb8
19. Nc6 Qc8
20. Nfe7+ Kh7
21. Nxc8 Bxc8
22. Ra8 Re8
23. Rxc8 g6
24. Rxd8 Rxd8
25. Nxd8 Kg7
26. f3 Nf8
27. Bxf6+ Kxf6
28. f4 exf4
29. h4 Ke5
30. Nxf7+ Kf6
31. Ng5 c5
32. dxc6 Ke5
33. Qc3#
[/pgn]
Also, something doesn't quite work with mates (or distance to mate). It just played 32... Ke5, accepting to be mated in 1, where it could easily have delayed the mate with any other move.
PS: It feels so good to finally be able to beat a computer. It won't last, so make the most of it.
Last edited by lucasart on Fri Mar 23, 2018 1:35 pm, edited 3 times in total.
Theory and practice sometimes clash. And when that happens, theory loses. Every single time.
jkiliani wrote:I think with a bit more strength, LCZero may become quite popular for people to play against, precisely because it doesn't play like a normal engine.
Judging from the experiences with Go, NN-based engines get strong at positional judgement much earlier than at tactics, which is kind of the opposite of Alpha-Beta engines (who can be tactical beasts but strategically clueless)
It might already be now if it gets more time. Here it only had 200ms/move, the guy who runs it estimated 200 playouts roughly. That really isn't a lot, with a larger search this might already be very interesting to face.