帖子的4楼说
I've had Leela 0.10.0 play five games against 0.11.0 Beta 1 a few days ago and the beta version only won three of those, so I'm not sure how the playout improvements affect strength.
Here is a game I just ran with the "beta 3" version. That's on a Linux server with 6 Intel Xeon X5650 2.67 GHz, 24 threads. The beta lost because it misread a life-and-death problem. (The 0.10.0 version's life-and-death behaviour is no less weird.) So the increased evaluation strength is not reflected in its life-and-death or semeai either.