1

In a blitz game between Leela zero and Komodo, the following position came up.

[title "Leela Zero vs Komodo"]
[fen "8/8/2p3k1/4bpp1/2P3P1/4BK1P/5P2/8 w - - 0 1"]

With White to move. Or WHITE King on f3 Bishop on e3 pawns on c4 f2 g4 h3 BLACK King on g6 Bishop on e5 pawns on c6 f5 g5. WHITE to move.

The game ended with a draw, although at a certain point both engines gave quite a high evaluation score. We know Leela is notoriously weak in the endgame. Did Leela zero miss something? What happens with perfect play?

2

2 Answers 2

2

Tried to run a match between stockfish 14.1 each at 2 threads with 6-men ending tablebase on TC 15m+5s.

Result

[Event "Engine match"]
[Site "?"]
[Date "2022.02.01"]
[Round "1"]
[White "Stockfish 14.1"]
[Black "Stockfish 14.1"]
[Result "1-0"]
[FEN "8/8/2p3k1/4bpp1/2P3P1/4BK1P/5P2/8 w - - 0 1"]
[SetUp "1"]
[TimeControl "900+5"]

1.Bb6 { [%emt 0:00:17] [%eval 308,60] } 1...Bd6 { [%emt 0:00:18] [%eval 341,72] } 2.Ke3 { [%emt 0:00:12] [%eval 312,51] } 2...Be5 { [%emt 0:00:09] [%eval 337,65] } 3.Bd8 { [%emt 0:00:23] [%eval 306,62] } 3...Bc3 { [%emt 0:00:12] [%eval 350,67] } 4.Bc7 { [%emt 0:00:09] [%eval 300,62] } 4...Kf6 { [%emt 0:00:11] [%eval 341,61] } 5.Kf3 { [%emt 0:00:22] [%eval 291,68] } 5...Bd4 { [%emt 0:00:16] [%eval 357,60] } 6.Kg3 { [%emt 0:00:11] [%eval 320,74] } 6...f4+ { [%emt 0:00:17] [%eval 371,61] } 7.Kf3 { [%emt 0:00:11] [%eval 548,65] } 7...Ke7 { [%emt 0:00:40] [%eval 555,65] } 8.h4 { [%emt 0:00:12] [%eval 548,54] } 8...Kd7 { [%emt 0:00:14] [%eval 555,53] } 9.Bb8 { [%emt 0:00:12] [%eval 551,54] } 9...Kc8 { [%emt 0:00:14] [%eval 559,56] } 10.hxg5 { [%emt 0:00:13] [%eval 850,49] } 10...Kxb8 { [%emt 0:00:50] [%eval 718,55] } 11.Kxf4 { [%emt 0:00:13] [%eval 914,44] } 11...Kc7 { [%emt 0:00:43] [%eval 865,47] } 12.Kf5 { [%emt 0:00:13] [%eval 957,42] } 12...Kd6 { [%emt 0:01:54] [%eval 1005,54] } 13.f4 { [%emt 0:00:15] [%eval 1047,39] } 13...Kc5 { [%emt 0:00:22] [%eval 1018,48] } 14.Ke6 { [%emt 0:00:14] [%eval 1284,44] } 14...Kxc4 { [%emt 0:00:29] [%eval 1538,43] } 15.f5 { [%emt 0:00:15] [%eval 1576,37] } 15...c5 { [%emt 0:00:53] [%eval 15257,36] } 16.g6 { [%emt 0:00:17] [%eval 15258,30] } 16...Kb4 { [%emt 0:00:19] [%eval 15260,38] } 17.f6 { [%emt 0:00:17] [%eval 15260,44] } 17...c4 { [%emt 0:00:11] [%eval 15261,40] } 18.g7 { [%emt 0:02:38] [%eval 32726,43] } 18...c3 { [%emt 0:00:12] [%eval 15262,44] } 19.g8=Q { [%emt 0:00:15] [%eval 32730,42] } 1-0


2
  • 6...f4? looks horrible though. What happens if Black just waits with 6...Bc5 or 6...Kg6 ?
    – Evargalo
    Feb 1, 2022 at 14:41
  • According to engine analysis, both can be answered by f4 still with advantage to white.
    – ferdy
    Feb 1, 2022 at 16:49
-2

Again , in the following endgame Leelazero vs Dragon WHITE King on g1 Bishop on d3 Knight on e2 pawns on a3 b2 d4 e5 f2 h3 BLACK King on g8 Knights on a7 c6 pawns on a4 b6 d5 e6 f5 h7 BLACK to move the endgame appeared to be winning for White. Analysis 40 minutes on Stockfish 14 NNUE 4 thread , with a second iteration at a critical point. It is confirmed Leela zero is far from infallible in the endgame. In the game, it terminated with a draw.

2
  • Are any current engines infallible in the endgame? (I think not) Feb 2, 2022 at 14:51
  • I didn't downvote but I do think this answer could be greatly improved by (1) forming a well-reasoned evidence-based argument and (2) adding the FEN of the position (or a link/moves to the game mentioned). Feb 2, 2022 at 14:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.