From the games, it seems evident that this adversarial system didn't find anything inherently Go-ish (for example, difficulty calculating complex ladders). Inst

From the games, it seems evident that this adversarial system didn't find anything inherently Go-ish (for example, difficulty calculating complex ladders). Instead, it appears that KataGo used in these has been taught with a different rule set and then makes a silly mistake when put in a situation with different rules. Here are some explanations for those of you who are not Go players. Ladders are structures where a specific repetitive pattern emerges. They are thought to be novice Go players who will be able to calculate them. On the other hand, the Nature paper about AlphaGo Zero learned to handle them later in training. More about ladders: https://senseis.xmp.net/?Ladder https://senseis.xmp.net/?Ladder Go rules are said to be simple, but in reality, calculating points in certain corner cases is difficult. Hence, there are different rules where the point calculation method differs. Typically the result is identical, but there are cases where this is not the case. More about rules here: https://senseis.xmp.net/?RulesOfGo https://senseis.xmp.net/?RulesOfGo