Sandhini Agarwal: We did discover that it generated certain undesirable outputs, but they were being all things that GPT-3.five also generates. So in terms of danger, as a exploration preview—for the reason that https://youtu.be/aGubCnA5Qyk