New answers tagged

64 votes
Accepted

Is "Adversarial Policies Beat Professional-Level Go AIs" simply wrong?

One of the authors of the paper here. It might shock you to find that I believe the paper was not "simply wrong". I do however think that we did a poor job explaining our evaluation setting, ...
11 votes

Is "Adversarial Policies Beat Professional-Level Go AIs" simply wrong?

In the Tromp–Taylor rules, the concept of dead stones is left out—all stones on the board at the end of the game are treated as being alive. The KataGo AI was trained primarily using the Tromp–Taylor ...
14 votes

Is "Adversarial Policies Beat Professional-Level Go AIs" simply wrong?

Regardless of the validity of this particular strategy, the general technique seems quite valid. They did not simply discover a bug in KataGo and then write a program to exploit the bug. Rather, they ...
5 votes

Is "Adversarial Policies Beat Professional-Level Go AIs" simply wrong?

I think you're right - this isn't much of a weakness in KataGo. It looks like there's a mismatch between the way KataGo plays, and the scoring method they're using. In Tromp-Taylor scoring, White ...
  • 3,090

Top 50 recent answers are included