News from AlphaZero

[Javier Ros]

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

[Spliffjiffer]

evolution is in full progress (as always)...im so amazed by these new things going on right now...its seems to be a revolution from my pov...
thank u very much for the link Javier

[chrisw]

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

Well, it’s an article/blog written off the backof a PR release from a publisher who doesn’t yet have the book, which is not yet actually written by some chess guys who apparently get to see some hundred of games that nobody else has seen, selectively provided them by Google DeepMind, using not known criteria. And some quotes about something more in that classic forward moving window “in a couple of weeks”.

Excuse me for being an old cynic.

[Ozymandias]

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

Well, it’s an article/blog written off the backof a PR release from a publisher who doesn’t yet have the book, which is not yet actually written by some chess guys who apparently get to see some hundred of games that nobody else has seen, selectively provided them by Google DeepMind, using not known criteria. And some quotes about something more in that classic forward moving window “in a couple of weeks”.

Excuse me for being an old cynic.

Love the summary.

[MikeB]

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

Well, it’s an article/blog written off the backof a PR release from a publisher who doesn’t yet have the book, which is not yet actually written by some chess guys who apparently get to see some hundred of games that nobody else has seen, selectively provided them by Google DeepMind, using not known criteria. And some quotes about something more in that classic forward moving window “in a couple of weeks”.

Excuse me for being an old cynic.

+1

[Nay Lin Tun]

New hype for new year, we may probably see new Alpha Zero version 2 in 2019.

They say " Future of chess, Revolution of chess".

[jp]

Well, it’s an article/blog written off the backof a PR release from a publisher who doesn’t yet have the book, which is not yet actually written by some chess guys who apparently get to see some hundred of games that nobody else has seen, selectively provided them by Google DeepMind, using not known criteria. And some quotes about something more in that classic forward moving window “in a couple of weeks”.

Excuse me for being an old cynic.

Not an old cynic, but too polite. We all know what the criteria are for the selectively provided games.
Authors should not allow themselves to be used like that.

[MikeGL]

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

Thanks for the link, i used Google translator and made me smile after reading "tower five farmers".
Maybe AlphaZero can double as a Google translator, but kidding aside the gist of the message is clear.


The value of figures, tower five farmers, knight three, it does not know.
The value of pieces, Rook five pawns, knight three, it does not know.


--------TRANSLATED FROM THE LINKED ARTICLE-------------------
...................................................................................................................................
The world's best source for detailed and deeply commentated AlphaZero games (in German and English) is probably the Bears of Lake Constance game database. That could change in March. Then the book "Game Changer" appears on the Google machine (see below).

For this flare-up of a new chess star, followed by its disappearance, we find a parallel in chess history, and it's set to end in 2017. The chess program AlphaZero only knew the rules when it began playing 40 million games against itself and others within nine hours play. After that it won a match against Stockfish, the best chess program until then. And was never spotted playing chess again.

Offside the parallel there is a significant difference. Sultan Khan was world-class right away, among the ten best players on the planet. AlphaZero was much better, and to a degree that gives its chess new insights into how the game runs when played close to perfection. AlphaZero has taught us that beyond the deep tactical chess of conventional machines, until then the measure of all things, there is one more level, the pursuit of positional dominance, regardless of material considerations.

Unlike Mir Sultan Khan, AlphaZero has re-emerged, and in the person of its creator. Demis Hassabis, head of Google firm DeepMind, which developed AlphaZero, was now a guest at the World Chess Championship. He performed the first move of the seventh game, after which he stood in the studio Judit Polgar and Anna Rudolf to speak and answer, but had to leave quickly.


What Caruana's goal was "1.e4" was easy to decipher for Demis Hassabis when performing the first move. The boss of Google's daughter DeepMind is a strong chess player. After the opening move, he chatted at the live show with Judit Polgar and Anna Rudolf from the AlphaZero sewing box. (Photo: Worldchess)

When the eighth game was due the next day, AlphaZero was twice as good. Demis Hassabis came back to the studio to answer all the questions, while AlphaZero's little sister Leela was on the board. With this Fabiano Caruana had prepared that Sicilian play, which caused Carlsen considerable headache and a position on the brink of loss. Now the audience saw the effects of the fact that at this World Cup match neural networks play on both sides for the first time.
Not even his creator understands how AlphaZero finds traits

When Caruana had reached the end of his preparation he was superior, but on his own he was unable to turn that advantage into a full point. And this despite the fact that the thinking processes of a neural network resemble those of humans rather than those of a conventional computing servant such as Stockfish. Like humans, AlphaZero follows patterns it has learned against itself over the course of matches, and is free of rigid rules.

Scarce, Magnus Carlsen was able to free himself from Leiel's stranglehold in the eighth match of the World Cup match, as on the other side of the board Fabiano Caruana at the end of his preparation was on his own. (Illustration: Willum Morsch)

Demis Hassabis explains: "AlphaZero rates everything in the context of the current position. The value of figures, tower five farmers, knight three, it does not know. That's why, for example, it's easier for him to sacrifice quality in the long term for positional reasons. Stockfish can do that as well, but the material balance of minus two forces it to be so profound that it sees a real advantage. "
Soon a new AlphaZero match?

However, even the creator of AlphaZero can not explain exactly how AlphaZero and Leela come up with their superhuman maneuvers. "For the moment this is still a kind of black box. We do not know. "But that should change. DeepMind plans to roll back AlphaZero's algorithms bit by bit to better understand how the decision was made
............................................................................................................

[peter]

Hi Mike!

It seems that soon we will see AlphaZero again in action!!

https://perlenvombodensee.wordpress.com ... -wm-match/

Thanks for the link, i used Google translator and made me smile after reading "tower five farmers".
Maybe AlphaZero can double as a Google translator, but kidding aside the gist of the message is clear.

For a native german speaker, the article is even more fun to read.

[whereagles]

you guys are just jealous you didn't program alphaZero