30: This seems less interesting than the headline makes it sound. It seems like they just fooled KataGo's algorithm for dealing with the end of the game, making it end the game early in a clearly winning position and lose on a technicality.
I think it would be interesting if you could find an actual way to win based on adversarial examples where you play weird moves and convince KataGo to make stupid responses, like in:
30: This seems less interesting than the headline makes it sound. It seems like they just fooled KataGo's algorithm for dealing with the end of the game, making it end the game early in a clearly winning position and lose on a technicality.
I think it would be interesting if you could find an actual way to win based on adversarial examples where you play weird moves and convince KataGo to make stupid responses, like in:
https://www.technologyreview.com/2020/02/28/905615/