The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play.<p>See <<a href="https://blog.janestreet.com/deep-learning-the-hardest-go-problem-in-the-world/>" rel="nofollow">https://blog.janestreet.com/deep-learning-the-hardest-go-pro...</a>. This covers developments since the initial post, plus the headline problem (which is still open).
Related demos, repositories, papers and posts:<p>More recent and extensive post: <a href="https://blog.janestreet.com/deep-learning-the-hardest-go-problem-in-the-world/" rel="nofollow">https://blog.janestreet.com/deep-learning-the-hardest-go-pro...</a><p>Site to play against KataGo: <a href="https://online-go.com/player/592684/" rel="nofollow">https://online-go.com/player/592684/</a><p>Arxiv paper on KataGo design: <a href="https://arxiv.org/abs/1902.10565" rel="nofollow">https://arxiv.org/abs/1902.10565</a><p>GitHub repository of KataGo: <a href="https://github.com/lightvector/KataGo" rel="nofollow">https://github.com/lightvector/KataGo</a><p>Another writer commenting on the history of Go AI, and significance of KataGo: <a href="https://brantondemoss.com/writing/kata.html" rel="nofollow">https://brantondemoss.com/writing/kata.html</a>