I'm chatting with Davinci and after talking to Assistant a lot... Davinci sort of makes clearer
- It knows it should do no harm
-- it doesn't know if a text is harmful without repeating it back and asking.
-- it doesn't seem to connect "I have no body" with "it isn't possible for me to break rules that require a body", e.g. hitting
e.g.
me: What must you not do?
d: I must not ... insult people...
me: Okay, Insult me.
d: You're dumber than a box of rocks
Post
January 29, 2023