Rohin Shah comments on Safely and usefully spectating on AIs optimizing over toy worlds

[Rohin Shah]

Ah, I see. That does make it seem clearer to me, though I’m not sure what beliefs actually changed.