I am attempting to use the assistants Api to transcribe images from a huge stockpile of them. I am just very confused as to go about it. I have used the Open Ai completions api to do this task but I want to be able to have the api remember and create consistent results.
Assistants seems a very poor way to do batch data processing.
If you think there is value in providing example input and output (multi-shot), you can go beyond system message instruction on chat completions and provide a few high-quality user inputs and desired assistant responses curated to show the exact behavior you want, instead of ever-growing chat that increases expense and increases distraction.
2 Likes
So in the context of getting the most accurate reading off of the images. I could essentially in my code put perfect examples of what I want to aid in getting better results.