zeshen comments on On thinking about AI risks concretely