My own opinion on Altman is the same as with GPT-2 getting a slow roll-out for safety: regardless of the reality, it is best to have a system and process which assumes the worst and can cope with that possibility being a reality.