The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play. See <https://blog.janestreet.com/deep-learning-t

The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play. See <https://blog.janestreet.com/deep-learning-the-hardest-go-problem-in-the-world/> https://blog.janestreet.com/deep-learning-the-hardest-go-pro.... This covers developments since the initial post, plus the headline problem (which is still open).