Komodo vs GM Smerdon knight odds match

[lkaufman]

Nah, giving away pieces for free because you view all options as certain losses is just a pathological side-effect. It's definitely not a way of playing less incorrectly

I'm not sure later Leela nets give away pieces more than earlier nets. Is there evidence they do?

Then again, I'm not sure later Leela nets don't hustle and fake as much as earlier nets either.

This is not something that requires testing. Version 11248 beat Naroditsky in blitz at about piece odds in a match. Any version starting with a 2 or higher number, if asked to play a game at knight odds, will start making silly blunders within a few moves. Even if you are rated 1500 or less, you can easily beat any of these versions at knight odds, it's totally obvious how awful they are.

[jp]

Any version starting with a 2 or higher number, if asked to play a game at knight odds, will start making silly blunders within a few moves.
... it's totally obvious how awful they are.

That's bad. I wonder if that makes Leela bad for analyzing unclear positions material down too. (Or, if it needs to be worse than unclear, positions down material but "with practical chances".)

[lkaufman]

Any version starting with a 2 or higher number, if asked to play a game at knight odds, will start making silly blunders within a few moves.
... it's totally obvious how awful they are.

That's bad. I wonder if that makes Leela bad for analyzing unclear positions material down too. (Or, if it needs to be worse than unclear, positions down material but "with practical chances".)

As long as Leela judges the win prob. to be above 5% or so, I think it plays ok, with the cutoff depending somewhat on the version. But once it gets close to what it considers resignable, it ceases to be able to tell good moves from bad ones. I noticed that the L14 network is a bit better than the regular Lc0 networks at knight odds in that it generally plays good positional moves, but still blunders pieces sometimes. Lc0 is still very strong at pawn and move (f7) odds, the win prob. is above the threshold, but much beyond that handicap it falls apart.

[jp]

We got a top commentator for it, former Ukraine champion GM Evgenij Miroshnichenko, now about 2600 FIDE, but nearly 2700 FIDE in 2009.

Miro predicts the human will win 5 1/2 - 1/2.

1.d4 e3.
Exchange French.
4. Nf3 after a long think by Komodo.
4. ... Nf6.

GM Svidler: "I do not understand how you score less than 5 points."

5. Bb5 c6 6. Bd3 Bd6. Miro thinks Komodo is playing badly.
7. 0-0 0-0. IM Toth thinks "game over" already.
8. h3 h6. Komodo is playing extremely slowly.
9. c3 Re8 10. Be3. They cannot understand how the human could not win.
10 ... Be6 11. Ne5 c5 12. Rc1 instantly.
12. ... cxd4 13. cxd4 Qb6 14. f4 Nc6 15. Qe2 Ne4?? 16. Bxe4 dxe4 17. d5 Qc7?= 18. Nxc6 bxc6 19. dxe6 Rxe6 20. Rc4 f5 21. Kh1 Kh7.

[jp]

22. Qc2. Be7 23. g4 g6 24. Rd4 Rd8 25. Qb3 Qc8 26. Rxd8 Bxd8 27. Bxa7 Bf6 28. gxf5 gxf5 29. Be3 Bg7 30. a4 Rd6 31. Qf7 Rf6 32. Qe7 Rg6 33. Rg1 Rxg1 34. Kxg1 Kg6 35. a5 Bxb2 36. Qd6+ Kh7 37. Qc5 Bf6 38. Qc4 c5 39. Qxc5 Qd7? 40. Qa7 +- Qxa7 41. Bxa7 Be7 42. a6

1-0. Komodo wins!

Ahem...

Larry, don't you think playing regular serious games is required for more than just keeping up with opening theory?

I'd think a GM who hasn't might be more prone to miscalculations, even hanging pieces.

15. ...Ne4??

[jp]

Game 2: French defence.

Komodo plays a3. Followed by b4. N trade on e4, opening the d file. Komodo trades Qs.

Komodo takes the e4 pawn.

Smerdon plays ... Nb8 instead of the better Nd8.

Komodo plays f5. But White has nothing and should lose this endgame.

[jp]

0-1. Komodo is checkmated.

Match level 1-1.

----------------------

Game 3: after 6 moves, already -5.00. Komodo needs a better opening book.
Miro says the opening went extremely well for Smerdon.

Smerdon plays h6 and g5.

After 15 moves, Black should win easily.

[Chessqueen]

0-1. Komodo is checkmated.

Match level 1-1.

----------------------

Game 3: after 6 moves, already -5.00. Komodo needs a better opening book.
Miro says the opening went extremely well for Smerdon.

I was analyzing with the Regular Komodo and it detected that Komodo MTCS made a horrible blunder. I Believe that you should use the regular Komodo for Tomorrow, and that will create more tension to the GM knowing that he will be playing with a totally different version of Komodo.

PS: Can they also use the other version of Komodo to analyze besides the commentaries of the Humans ?

[Chessqueen]

0-1. Komodo is checkmated.

Match level 1-1.

----------------------

Game 3: after 6 moves, already -5.00. Komodo needs a better opening book.
Miro says the opening went extremely well for Smerdon.

I was analyzing with the Regular Komodo and it detected that Komodo MTCS made a horrible blunder. I Believe that you should use the regular Komodo for Tomorrow, and that will create more tension to the GM knowing that he will be playing with a totally different version of Komodo.

PS: Can they also use the other version of Komodo to analyze besides the commentaries of the Humans ?

Komodo MCTS best chance is to play a Knight for a pawn, whereas the GM will remove one of his pawn and the engine his knight. Probably in 6 months Komodo NN might beat the same GM with a Knight Odd. You can probably raise the price to $150 for a Win and $50 for a Draw, if the GM accept to play the rest of the games tomorrow with a Knight for a pawn.

[lkaufman]

0-1. Komodo is checkmated.

Match level 1-1.

----------------------

Game 3: after 6 moves, already -5.00. Komodo needs a better opening book.
Miro says the opening went extremely well for Smerdon.

I was analyzing with the Regular Komodo and it detected that Komodo MTCS made a horrible blunder. I Believe that you should use the regular Komodo for Tomorrow, and that will create more tension to the GM knowing that he will be playing with a totally different version of Komodo.

PS: Can they also use the other version of Komodo to analyze besides the commentaries of the Humans ?

Komodo MCTS best chance is to play a Knight for a pawn, whereas the GM will remove one of his pawn and the engine his knight. Probably in 6 months Komodo NN might beat the same GM with a Knight Odd. You can probably raise the price to $150 for a Win and $50 for a Draw, if the GM accept to play the rest of the games tomorrow with a Knight for a pawn.

Komodo (both versions) already scored 2.5 out of 3 giving knight for pawn (details differed) to MVL and Naka, perhaps the two best rapid players in the world after MC. For a future match, GM Alex Lenderman suggested knight for tempo; White plays his move, Black uses his first move to "capture" the knight (remove it), and White gets a second move, presumably 1.e4 and 2.d4, then it's Black's turn. Philidor used to give this handicap. I think it is suitable for a midrange GM in rapid. It just needs a good name!