I’m no fan of “ai safety” but plenty in this post that was interesting for other reasons