lichess.org
Donate

Alphazero x SF: Why can we see only 10 games?

Will we ever see the other 90 games? I made a search and found nothing that indicated an answer to this question. We are talking about no less than 18 wins against SF, and 72 draws. All hidden. Why? What are they afraid of?
What was published on arXiv is a preprint.

Including all 100 games would take a lot of space for what's supposed to be a short preprint.

Also, even if they published the 100, then people would be like "Why not publish the 1200 games they played against SF from the 12 set opening positions?" and then "Why not publish all 44 million training games it played against itself?" (and actually people have already complained about both of those anyway).

They had to pick something (if they wanted to include anything), and picking 10 spectacular games that highlight the new approach's strength over traditional approaches makes sense.

Assuming that they're hiding something when there are much more mundane explanations is just a bit too conspiratorial for my taste.

I'll wait for the full paper before complaining about what was and wasn't included in the preprint.

Also, no games against Elmo were published. Where's the outcry about that? :)
Google is controlled by EA and they want us to experience a sense of pride and accomplishment while unlocking new games.

Serious: It would be really good if they published all 44 Million games as a Database. We would have plenty of new theory.
I'd also like to see those 1200 games.
If I develop an AI and teach it chess in the future, I'll publish all its training games.

Also: this kind of AI would be good for solving Atomic .

This topic has been archived and can no longer be replied to.