I am trying to create a GPT-3 chatbot that can read data from an Excel and answer questions in a natural way.
The Excel contains information about products in stock, their availability in different locations, the products that replace them and the prices of each one.
I would like users to be able to ask questions in a natural way, such as “What products are available at my local store?” or “What is the price of the X model?” and that the chatbot can find the answer in Excel.
Does anyone know how I can make GPT-3 just look up information in the Excel and respond in a coherent and natural way?
I am programming it in python but I didnt find a similar example yet.
Is there a GPT-3 specific function that can help me with this? Any help or advice is greatly appreciated.
You’re talking about having a program read a CSV file (what Excel reads) and then either generate or respond to prompts based on data in the CSV. This could become expensive extremely quickly, because you don’t want to be paying for GPT to read through 2 million entries. You’re only afforded so much memory anyways, this wouldn’t be possible.
I can imagine two methods to make this thing somewhat workable:
have a basic search feature for your database with the ability to query GPT after the relevant data has been retrieved.
you do something absolutely insane and let GPT write queries for your database based on an approximation of user input and it then replies with the information fetched from queries.
Option 2 is almost guaranteed to be a disaster, but it would be interesting.