1

I can break my puzzle down into the following:

  • How strong is Leela Zero today in terms of rating / strength, esp when compared with Alpha Zero / Mu Zero?
  • Does its rating get calculated as it continuously trains?
  • Will there be matches between Leela Zero and the other AI to derive a reliable rating?
3

We can't compare Leela vs. AlphaZero or MuZero, because they are private engines. That means we can't just get them to play tens of thousands of games against each other (which is the standard way of telling how strong each engine is). However, if AlphaZero has not improved since it was unveiled, Leela is likely to be stronger than it. The same goes for MuZero, since MuZero only got as strong as AlphaZero. The reasoning is the same as my answer to this question.

Leela has a self-play elo as it trains. You can see its progress here.

Leela already plays lots of games against other chess engines. You can watch its tournament games here, see its tournament results on Wikipedia, and see its position in rating lists such as CCRL and SP-CC.de. Caveat: if you are really interested in how "good" Leela is at chess, be sure to check out what the hardware used is.

3

How strong is Leela Zero today in terms of rating / strength, esp when compared with Alpha Zero / Mu Zero?

We will never know for sure because we don't have the source code for AlphaZero/MuZero. However, it's very likely Leela Zero is a stronger version because it is an evolving project. Open source for the best AI programmers in the world, not just internally to Google.

Will there be matches between Leela Zero and the other AI to derive a reliable rating?

A lot of computer tournaments are running. Check TCEC, CCRL etc. You could run your own tournament easily, it's free and open source!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.