p.b. comments on METR: Measuring AI Ability to Complete Long Tasks