Sandhini Agarwal: We did find that it generated selected unwanted outputs, However they were being all things that GPT-3.5 also generates. So in terms of danger, like a exploration preview—for the reason that that’s https://youtu.be/aGubCnA5Qyk