Daniel Kokotajlo comments on Forecasting thread: How does AI risk level vary based on timelines?