Question answer about Products

Prompting
1

hi, i am thinking of designing a question answer for bunch of products based on product data i have. e.g. file with all the products could look like this (open to suggestions here):
{‘text’:‘Product A can train your dog to go to a specific place. Price of the product is $10. …’,‘metadata’:‘product A’}
{‘text’:‘Product B can train your dog to go to find a specific object. Price of the product is $20. …’,‘metadata’:‘product B’}

Now i am hoping to answers user questions like this:
Human User: Can i train my dog to go to specific place.
Answer. Yes. Use Product A (i can look at semantic search and use selected document and its metadata to generate this answer if i want to control it)
Human: how much does it cost?
Answer: $20.

i have been playing with the api and it seem to be working reasonably well for answering the first question. it even shows the document correctly based on which it is answering question. Next time the question is gettting asked, how can i ask it to look in a specific document within the file. when i am using documents parameter, its giving me BS answer :slight_smile:

any pointer will be appreciated?

3

Hey, The kind of solution you are hoping for, where Open AI sort of “learns” all the data from a document and answers questions, is complicated. In fact Open AI itself is still pretty bad when it comes to this, you will have to do a lot of fine tuning and prompt engineering to get there.

Instead try this https://github.com/openai/openai-cookbook/tree/main/examples/fine-tuned_qa