Here is some interesting data from the Leela Network. Page 23 from the supplementary data sheet of the AlphaZero Paper is also quite stunning. For example it tells us that this is a forced draw:

To put it mildly, I am not entirely convinced. There is a similarity to using an engine in correspondence chess. If you just play the top recommendation you won’t get far. It is well known that engines, even neuronal nets, have horizon issues that can be exploited.