points
Very interesting that he seems to be in the camp of “It’s ok if the machines prove it as long as we can understand and formally verify it after.”