Hello everyone, is it possible to use the files capability as a database for examples? In a way where it doesn’t just use those examples but also the entire training data like it would in a normal prompt.
Thank you very much.
PS: I LOVE THIS! It blows my mind to shreds every day.
I’m not the OP, but I’m wondering if their idea is similar to mine.
I was considering whether GPT-3 could write a research report. So, for example, I would supply an example of an output (research report) and GPT-3 could use that as well as its training data to generate output.
My problem is the size of the prompt and output. Based on the developer guidelines and current token limits, I don’t think this would work right now.
Thanks for the response and kind words! 
I think I am missing how to input files to the API to begin with (I can barely code, so I need the absolute idiot’s version.) Let me look at the documentation again and see if I can understand how to do this. I may have missed it.
I’ve been experimenting with using one instance of the model to summarize a longer doc, then pass that summary to the new instance so that it can continue to predict. I’ve had mild success with that on my trials.