Quoting Jakub Pachocki

The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI. [...] We will …