Leela Zero( A Neural Network engine similar to Alpha Zero)

That win is one of the most amazing things I have ever seen.

Its strange seeing these people crumble when playing Leela. You'd figure they wouldnt get turned into bumblers....but there must be something about playing Leela in public. Id be interested in the elo rating of Tang in the match....probably about 300-400 less than usual?

1st game(15 min) - 37 centipawn loss, 5 inaccurate, 3 mistakes, 0 blunders.

2nd(15 min) - 33 centipawn loss, 7 inaccurate, 2 mistakes.

3rd (5 min) - 50 centipawn loss, 5 inaccurate, 7 mistakes...thats 1/3 of the moves.

4th(5 min) - 48 centipawn loss, 9 inaccurate, 1 mistake, 1 blunder....38 moves.

Those Centi pawn loss are grossly incorrect assessment of game play.

Let me give you an example, I downloaded a high quality stockfish play from  TCEC and let that  stockfish assess there , there was as high as 0.13 centi pawn loss. Why? there are discrepancies in moves between high and low ply. 

This is my last game in lichess account, my rating is 2100+ in bullet, the average centi pawn loss was 0.24. Does that mean , am I better than Andrew? It is useful to checkout blunders only.

https://lichess.org/5oAU9PxLKd1j

I can play <0.20 centi pawn loss in some solid positional opening such as QID, but my game play can be as worse as 0.50 in highly tactical opening such as sicilian.

Just an approximation.

I did play against Leela as white and lost. 55 centipawn loss thanks to a huge silly blunder. There was no wiggle room....with a meager 9 centipawn loss for it from the French defense Nc3/Nf6.

I'll let others play hehehe.....seems to be a long line to play!

Wow!

Leela is stronger than Stockfish 1.01, now!! And now, Stockfish 9 and 10 are waiting.

Running the cpu version of Leela against Monarch 1.7, ID 194 is definitely significantly stronger than ID 178. And, Leela seems to be much less stupid in the endgame. :)

which win?

click the highlighted blue link

Another article on the match with GM Tang is here: https://lichess.org/blog/WtzZAyoAALvE8ZSQ/gm-andrew-tang-defends-humanity-against-leela-chess-zero

https://en.chessbase.com/post/leela-chess-zero-alphazero-for-the-pc#

Chessbase Article

https://www.gofundme.com/leela-chess-zero

Official funding!

Leela's first draw vs SF 10 dev version on my machine.

I did a 2/1 blitz tourney Leela ID 202 vs stockfish dev version around April 12-15 ,on my i5 7400 3Gh, Nvidia 1060 3GB. I did full tuning of GPU, and got 2k-3k nps average nps for Leela, meanwhile  stockfish dev version got about 8000 knps.SF is on  4 cores full strength! 

Drmr, thank you for these great news. I just downloaded the dev version havn't tried it yet. Is it stronger than SF9? SF9 is pretty new so I doubt this dev version could be really better (?)

Have you tried Leela against SF9?

SF  10 dev version is about 20 elo stronger than SF 9.  Current Leela is 2800-3000 elo in my 1060 Nvidia, SF dev 10 version is 3400+ elo. 

Leela opening choice (probability of winning ) from initial position!

Leela's White 1st Move Preferences:

e4 > Nf3 > d4 > c4 > g3 > e3 > Nc3 > b3 > h3 > d3> c3 > a3 > f4 > a4 > b4 > Nh3 > Na3 > f3 > g4

Very interesting!  And mostly makes sense, maybe except for 1.g4 in last place.  And maybe 1.f4 should come before 1.h3 and 1.a3.

Leela id 200 got checkmated and id 215 draw 

https://lichess.org/forum/game-analysis/checkmated-leela-2700-id200

https://lichess.org/forum/game-analysis/draw-vs-leela-2700-id-215

Oh i see its 2800 rated. Nice.

Recent ID  210 CCRL rating =2885, There is plan to increase network size from 10x124 to 15x192 but surprisingly the larger beta network doesnt increase elo.