gwern comments on Notes on the Safety in Artificial Intelligence conference