Is is possible to use the fine tune option of OpenAI and use the outputs of previous prompt-completions (generated via the API) with a grading to indicate how good these were? So, in this case it will not provide the ‘ideal completion’ but only a grading.
I don’t believe this is possible, right now, the current fine tuning API requires that you provide a prompt and associated completion as the data to fine tune on. It would be great if you could grade previous responses but this does not exist today.