Extended data figure 2 contains something really cool: how long it takes for AG0 to discover a joseki, and how long it takes it to later discard it as non-jokek

Extended data figure 2 contains something really cool: how long it takes for AG0 to discover a joseki, and how long it takes it to later discard it as non-jokeki. So you could, in theory, evaluate a human joseki by plotting AG0's probability of playing it against training time (or against its Elo rating). What's also cool is that the differences that cause it stop playing a joseki must be really minute, but it can see them anyway.