rorygreig comments on Talk: “AI Would Be A Lot Less Alarming If We Understood Agents”