It's a neat idea, but saying that this AI "mastered" chess is a huge exaggeration. In the actual paper [1], it says that it lost 19/100 games against an opponen

It's a neat idea, but saying that this AI "mastered" chess is a huge exaggeration. In the actual paper [1], it says that it lost 19/100 games against an opponent playing completely random moves. And by "lost", they mean that it was in a worse position after 40 moves. It sounds like the AI isn't even able to consistently finish games, so they just had to try to judge them at that point. [1]: https://arxiv.org/pdf/1907.08321v1.pdf https://arxiv.org/pdf/1907.08321v1.pdf