How do I tell my davinci what it has to be/how it should act?

Prompting
1

I was watching a lot of videos and going through articles but can’t really tell what I should do.
I am building a QnA Chatbot with nodejs.
I prepared a qna.jsonl which has multiple lines that look like this:
{“prompt”: “<question>?\n\n###\n\n”, “completion”: " <answer> END"}

So how do I tell it smth like:
‘I am a highly intelligent question answering bot. If you ask me a question that is rooted in truth, I will give you the answer. If you ask me a question that is nonsense, trickery, or has no clear answer, I will respond with "Unknown".’
This is copied by the Playground. But how do I get it into my jsonl-File if that even belongs there?
Or has the system behaviour to be in every prompt. Because normally there is a q and an a in my line, but this is more like an advice without an answer and I don’t want that the model thinks this is supposed to be a question that someone asks. I know that it is possible to append that before the question by the user in the prompt. But when I concat that String infront of every message thats so repetitive and costs tokens each time even when someone would write “ok, thx” and I don’t even know if the model knows its and advice and not a something the user wants.

Also when I train my model again, would I add the new lines to the existing file and fine-tune it again or do I create a new one and it adds that to the past tuning.

Thanks for potential help.

2

You’d just send it as the prompt to the fine-tuned model.

If you’re adding data, you have to fine-tune the whole thing again.

As far as I know, a lot of people are moving away from fine-tuning and using embedding + newer models.

Are you just building a chat bot? IS there a reason you want to fine-tune?