HoldenKarnofsky comments on AI Safety Seems Hard to Measure