Leela Zero (Stockfish without human knowledge), like Alpha 0

Btw, if you would like to contribute in testing the Leela, I can help you set up your machine, we need a lot of volunteers. (Same as Stockfish project)

Nice. g4 was a surprise that turned out well!

But what was it up to at the end?  I believe this could be fixed by encouraging it to win quickly in the design of the reward calculation. This is very straightforward if it is only told about the value of a draw or a loss: score a win on move N as gamma^N, a loss on move N as -gamma^N and a draw as zero. where gamma is (of course) less than 1,

I know little about the details of the  design of the system, but it certainly looks as though it gives no value to winning more quickly. Any discounting would nudge it towards being efficient, which might also provide a better return on computational resources for learning and playing.

A true AI wont be told what a "spike" is, besides the outcome of the game. Humans telling the computer what a spike is, thats SF..or Komodo...or Houdini. Sure, with is 80,000 legit SF9 games already in, it will get to 3400 once it reaches those positions....they will boast how good it is, but it wont beat up on Komodo.So then they will add 100,000 Komodo games, and it becomes SF-Komodo and they boast again. Then they add 100,000 Houdini games and it will be SF-Komodo-Houdini...which will be impressive. But not more so than AlphaZero. With a new higher elo of any of those it will have to play again 100,000 games.

 Presumably Leela was named after the Futurama character?

Leela means divine play

I am starting to lose Leela,  lose several games on time on 3'+2" min blitz.  

[Event "Computer chess game"]
[Site "DESKTOP-U9VVTOA"]
[Date "2018.03.23"]
[Round "?"]
[White "Nay Lin Tun"]
[Black "Play

[001]

"]
[Result "1-0"]
[BlackElo "2000"]
[ECO "A07"]
[Opening "Reti Opening"]
[Time "23:16:36"]
[Variation "KIA, 2.g3"]
[WhiteElo "2400"]
[TimeControl "180+2"]
[Termination "normal"]
[PlyCount "111"]
[WhiteType "human"]
[BlackType "program"

Leela running on my Arena 3.5.1 GUI.

Sorry, I checked wrong eval, it was not Leela who saw M5, M1  , it was stockfish. But  Leela will fix by its own

I pointed out earlier (#27) that the odd characteristic of not trying to finish a game off could surely be fixed with a slight change to the reward function, and this might even be beneficial to performance.

Leela progress!

We got 100 active contributors, 76k games were played  3500+ elo increased from random play. (No one know the initial elo of random play, it might be arbitrarily -2000 elo).

Estimated elo of Leela=1800 now.

@Elroch, thanks for your feedback. It would be a good idea to discuss with Leela  programmers. Here is a link if you would like to discuss directly https://groups.google.com/forum/#!forum/lczero

Thanks for the suggestion. As you may have guessed, I have done spent quite a lot of time on related technology for other applications.

Leela is growing very fast. A lot of people start crying that they could not win him anymore now. For comparison

Leela has been trained for 838k games.

Alpha 0 has been trained for 42 million games.

Stockfish has been tested for 705 million games. 

how do we play her?

http://play.lczero.org

My first game against Leela with Lichess annotation using Stockfish 9.

Im noticing she plays exceptionally bad once the position is worse or lost for her.

Ok, I am pretty sure that Leela will be 2000+ in next month and  you wont win her anymore! (She is just one month old baby! )