TalkChess.com

I have feeling, that DeepMind is now ashamed of premature report and we get nothing new!

You have a feeling. We have 10 amazing games.

And 90 dreadful ones they didn't show

Why would you think that?

It's far more likely that the peer-reviewed paper they announced is stuck somewhere in the review process. If you ever published scientifically, you'll know how long it takes from the initial submission until your article actually comes out, especially if the reviewers have a lot of comments that take work to accommodate.

I do think Deepmind is rather anxious by now for the paper to finally come out since if LCZero was already strong by the time that happened, people would be a lot less excited.

How people immediatly jump to conclusions like that completly astonishes me. Do you actually like such conspiracy theories ? I dont get it, pls explain

I understand we live in a world where falsehood is common.
In my country we can't even trust on banks anymore (several curruption cases).
But something must be true right?
DeepMind looks truthfull/trusty.
For once the games they made available.
Also the pre-paper they published looks very convincing.
Also the hardware they used makes this whole thing credible.

Cardoso wrote:I understand we live in a world where falsehood is common.
In my country we can't even trust on banks anymore (several curruption cases).
But something must be true right?
DeepMind looks truthfull/trusty.
For once the games they made available.
Also the pre-paper they published looks very convincing.
Also the hardware they used makes this whole thing credible.

If Garry Kasparov went to play some Go games and then give himself a rating above current top players and title himself GM in Go based on self-play and some non-official games played against opponents he chose, there is something to believe, both are strategy games and he is indeed very clever, also this and also that, then do you think everything else he said must be true based on who he is?

Werewolf wrote:And 90 dreadful ones they didn't show

You are contradicting yourself. You say 90 games are dreadful, but you haven't even seen them.
How dreadful can be the other 18 wins or the 72 draws against SF8?
I agree it wasn't the latest and greatest SF version, and the engine settings for SF could be better. To me any win/draw against SF8 is anything but dreadfull. The difference in ELO between the latest SF9 and SF8 can't be "applied" to alpha zero, because SF and alpha zero are immensely different.
The SF testing framework is allways against the latest SF with all it's strengths and weaknesses.
Alpha zero aproach seems much more solid and weakness proof.

Cardoso wrote: The SF testing framework is allways against the latest SF with all it's strengths and weaknesses.
Alpha zero aproach seems much more solid and weakness proof.

Last time I checked the A0's training is solely based on self-play which also applies to the first statement.
Then I cannot reason how you come to the conclusion that one is more solid and weakness proof while the other is not.