[benchaney]

Not in Tromp Taylor rules. Basically, this paper is cheesing a win with a loophole in a rule set that no one actually uses. Kind of a shame that it is getting so much attention despite the fact that the result is so heavily exaggerated, and the actual accomplishment is pretty minimal.

[Tarsul]

Still, even as a hobby go player i think the broader point that AI usually only works in the constraints that it is designed, bears repeating. Because if journalists/scientists (and us) wouldn't do it, people would maybe start believing the marketing messages about AI too much. But yes, there are better examples (e.g. those delivery robots that don't work in real life conditions).