Jan Leike: We did carry out some more “crimson-teaming” for ChatGPT, in which All people at OpenAI sat down and experimented with to break the product. And we had exterior teams executing exactly the same type of matter. We also experienced an early-entry method with trustworthy users, who gave responses. https://elliottovaeh.verybigblog.com/31071439/details-fiction-and-chatgpt