Is it static? If I ask it the same question I asked it a few months ago it gave me a completely different answer. Is that because of some additional context above? Should we be starting fresh chats sooner?
However, the model is static (it’ll present the same candidate word list until retrained), which is what you may have heard. But the way a response is generated introduces the randomness.
Note: I forget the parameter, but if you get the direct API access you can turn off this and get consistent answers.
With a low enough temperature you get essentially the same output every time, with a at most just minor words swapped.