Alpha Zero vs. Leela Zero

[h1a8]

If Leela had access to the same number of games for training as Alpha Zero did then would it be as strong as the Alpha Zero on comparable hardware (something that matches 4 tpu).?

If this is unknown then
Maybe someone can run a 100 game match against stockfish 8 (with the same configuration as alpha zero played against). Leela uses hardware comparable to 4 tpu.

[dkappe]

Been there, done that, first with the t10 nets, then the t30. If you search the forum, you’ll even find a discussion about this topic.

Here a wiki page from November, 2018 declaring victory over sf8. https://github.com/dkappe/leela-chess-w ... ct-Success

[supersharp77]

Been there, done that, first with the t10 nets, then the t30. If you search the forum, you’ll even find a discussion about this topic.

Here a wiki page from November, 2018 declaring victory over sf8. https://github.com/dkappe/leela-chess-w ... ct-Success

Reviewed the game on youtube https://www.youtube.com/watch?v=xu037yC ... e=youtu.be
Stockfish loses the thread of a closed French opening..An opening Stockfish never plays in my tourneys..Stockfish avoids almost all closed position like a plague...I keep seeing these closed French lines on youtube though.
Anything from Mr 'Kingscrusher' I take with a grain of Salt ...this man has made a career out of Hyping first Alpha Zero and now Leela0 He posts his touts almost daily with the same theme..checked his site and he has posted hundreds and hundreds
of LC0 'Masterpieces'.. Had to look really hard to find any Stockfish victories he posted..I assume his views are higher for his Leela posts so the man keeps...Going and Going...and Going..

[mwyoung]

If Leela had access to the same number of games for training as Alpha Zero did then would it be as strong as the Alpha Zero on comparable hardware (something that matches 4 tpu).?

If this is unknown then
Maybe someone can run a 100 game match against stockfish 8 (with the same configuration as alpha zero played against). Leela uses hardware comparable to 4 tpu.

All the game data for the rating of AZ suggests that Stockfish 10 is stronger then AZ. And we know that Lc0 is stronger then Stockfish 10.
We can not play them head to head. But elo suggests that Lc0 is stronger then AZ.

But AZ could have played stronger with more TPU's and Training. But most Likely Lc0 is stronger then AZ that Played Stockfish 8.

[jp]

Anything from Mr 'Kingscrusher' I take with a grain of Salt ...this man has made a career out of Hyping first Alpha Zero and now Leela0

Yes, he's lost all objectivity. It's previously been suggested that tests be made of moves by different engines to see whether humans can tell which are AB-engine moves and which are NN-engine moves. I think he and everyone else would fail such a test, assuming the tests are series of good middlegame moves.

[Albert Silver]

If Leela had access to the same number of games for training as Alpha Zero did then would it be as strong as the Alpha Zero on comparable hardware (something that matches 4 tpu)?

Leela has had far more in every one of its runs. The current Leela that played and defeated Stockfish in TCEC was trained with well over 80 million games. The one that barely lost to Stockfish in the previous TCEC was trained on over 60 million games. AlphaZero was trained on 44 million.

Unless I misread your statement and you meant 'if Leela were limited to only 44 million games like AlphaZero'.

[Dann Corbit]

I would love to see alpha zero verses leela zero with unlimited hardware (e.g alpha zero with 100 TPU cards funded by google verses leela zero with 100 top GPU cards funded by Nvidia).
But if wishes were horses then beggars would ride.

[Ovyron]

Anything from Mr 'Kingscrusher' I take with a grain of Salt ...this man has made a career out of Hyping first Alpha Zero and now Leela0

This is absolutely false.

First, if you take a look at the views of his videos, it can be argued that this man has made a career out of playing tournaments at lichess.com.

Second, it can be argued that even before playing those tournaments at lichess, this man had already a career.

I was there when it happened, and the only reason he started focusing on Alpha, and then on Leela, was because the huge audience he already had was begging him for more Alpha and Leela videos. And it reached the point where his videos without Alpha/Leela on them didn't have much views on comparison. This was a supply/demand thing, people were really hungry to see analsys of the latest games from the NNs, Kingscrusher was providing great analysis of those games, and as his audience was racking up the views, new members joined to watch his channel, and they wanted to see more games by Leela, so all he did was keeping everyone happy.

This started at the very beginning of Leela's progress, back when you could go to their website, play agaisnt Leela on "Hard", and still beat Leela. Back then 99% of engine games played by Leela were lost by Leela, against other weak chess engine, so the news was when Leela won those games, and the unique way she was doing it (playing chess unlike anything that was seen before, unlike chess engines or humans, it was like chess from aliens from another dimension). Games where Leela lost were uninteresting, because you could beat her yourself on the website.

Nowadays, Kingscrusher reports Leela's loses, but he's getting his games from TCEC, so if Leela doesn't lose any game at TCEC, what else do you want him to report? He tried analyzing draws, but again, low views...

Because if you analyze a Magnus Carlsen game and get 1500 views and you report a Leela game and get 20000 views, what are you going to do next? Kingscrusher played a paramount role in getting Leela to be known by the masses, and getting them to see analysis of computer chess games. Before all this happened nobody cared about chess engine v chess engine games, entire Youtube channels with such videos died because nobody was watching them. Now people watching analysis videos of chess engines is a normal think, and we have to thank the people from DeepMind, the Leela Zero authors, and youtubers like Kingscrusher for that, because computer chess is no longer the small niche it used to be.

[mwyoung]

Anything from Mr 'Kingscrusher' I take with a grain of Salt ...this man has made a career out of Hyping first Alpha Zero and now Leela0

This is absolutely false.

First, if you take a look at the views of his videos, it can be argued that this man has made a career out of playing tournaments at lichess.com.

Second, it can be argued that even before playing those tournaments at lichess, this man had already a career.

I was there when it happened, and the only reason he started focusing on Alpha, and then on Leela, was because the huge audience he already had was begging him for more Alpha and Leela videos. And it reached the point where his videos without Alpha/Leela on them didn't have much views on comparison. This was a supply/demand thing, people were really hungry to see analsys of the latest games from the NNs, Kingscrusher was providing great analysis of those games, and as his audience was racking up the views, new members joined to watch his channel, and they wanted to see more games by Leela, so all he did was keeping everyone happy.

This started at the very beginning of Leela's progress, back when you could go to their website, play agaisnt Leela on "Hard", and still beat Leela. Back then 99% of engine games played by Leela were lost by Leela, against other weak chess engine, so the news was when Leela won those games, and the unique way she was doing it (playing chess unlike anything that was seen before, unlike chess engines or humans, it was like chess from aliens from another dimension). Games where Leela lost were uninteresting, because you could beat her yourself on the website.

Nowadays, Kingscrusher reports Leela's loses, but he's getting his games from TCEC, so if Leela doesn't lose any game at TCEC, what else do you want him to report? He tried analyzing draws, but again, low views...

Because if you analyze a Magnus Carlsen game and get 1500 views and you report a Leela game and get 20000 views, what are you going to do next? Kingscrusher played a paramount role in getting Leela to be known by the masses, and getting them to see analysis of computer chess games. Before all this happened nobody cared about chess engine v chess engine games, entire Youtube channels with such videos died because nobody was watching them. Now people watching analysis videos of chess engines is a normal think, and we have to thank the people from DeepMind, the Leela Zero authors, and youtubers like Kingscrusher for that, because computer chess is no longer the small niche it used to be.

Kingscrusher has been covering computer chess for years. Kingscrusher was not very chess engine savvy years ago. Here is a message I sent him over 5 years ago. When he did not know what Stockfish was and thought Houdini was still the best. This is before Stockfish won TCEC.

Mark Young
5 years ago
Thanks for the video. Stockfish has been the #1 chess engine in the world for many month now. Stockfish's rise to the top of the computer chess engine world may have gone unnoticed by some. Since its rise to the top was very rapid, and if you are not a computer chess engine tester. It would have been missed.

Stockfish is a free chess engine. And updates to the strongest version of the program are released almost daily. It supports the new table base standard in computer chess Syzygy Bases.

Latest and strongest version of stockfish can be found here.
http://abrok.eu/stockfish/

Stockfish with Syzygy Bases support can be found here.
http://abrok.eu/stockfish_syzygy/

Stockfish is a must have for any chess player.