Hello,
I’ve noticed that no matter how short the /completions response is (tried with 100-1000 tokens response), it never remembers/takes into account the previous question, so it doesn’t look like a conversation at all.
Any ideas about it?
Could you provide examples of your inputs and the corresponding outputs?
Strangely, (and sucks to make assumptions as I did) the davinci bot won’t use a chat ID to remember a conversational flow. The only approach I’ve found thus far is to post the latest of a previous chat as “History”.
Well firstly, Davinci is not a bot, its an engine. It can remember things that were said in the same context window, so if you want it to remember anything in any API call you have to remember it in the same prompt, which may be uncomfortable, all of this is because the 4k tokens limit per request.
Thanks for the smacked wrist I’ll note that down. And for providing the exact same response as I just did. Are you Davinci?