I suppose it’s possible OAI / MS had a good understanding of Bing Chat’s misbehaviors and decided that a closed beta was the best way forwards, but it seems unlikely to me.
...Why? A closed beta is a closed beta for this exact specific reason, because it’s kind of useful but you need a small group of users to test it and give you feedback first.
Seems like Bing Chat is at the stage where you hire contractors to interact with the AI, and have them agree not to post screenshots of the interaction on Reddit / Twitter. This avoids the reputational risk associated with putting your name on a product that insults, lies to and manipulates its users, and also avoids the risk of people pigeonholing Microsoft’s LM-agumented search products as “that insane offshoot of ChatGPT”.
...Why? A closed beta is a closed beta for this exact specific reason, because it’s kind of useful but you need a small group of users to test it and give you feedback first.
Seems like Bing Chat is at the stage where you hire contractors to interact with the AI, and have them agree not to post screenshots of the interaction on Reddit / Twitter. This avoids the reputational risk associated with putting your name on a product that insults, lies to and manipulates its users, and also avoids the risk of people pigeonholing Microsoft’s LM-agumented search products as “that insane offshoot of ChatGPT”.
Mainstream advice for launching new software products is to release quickly and then rapidly iterate based on user feedback.
See also: https://www.lesswrong.com/posts/jtoPawEhLNXNxvgTT/bing-chat-is-blatantly-aggressively-misaligned?commentId=xXcqbXPbGWwcqLskB