Tom Davidson comments on A framework for thinking about AI power-seeking