Sandhini Agarwal: We did see that it produced specific unwanted outputs, but they were all things which GPT-three.5 also generates. So with regard to threat, as a study preview—simply because that’s what it had been https://youtu.be/aGubCnA5Qyk