Sorry a basic question here as I am new to AI integration. I’m a product manager at at e-commerce company and trying to understand the feasibility of integrating ChatGPT into our search engine. We have a search engine that searches our “products, offers, etc” but are limited by basic keyword searches. Is it feasible to integrate ChatGPT on top of our knowledge base, train it, and have our users search through smart prompts to navigate and query various objects within our site? Would love to get some feedback if possible, thank you!
Small technical detail, but you can’t integrate ChatGPT into your site, but you can integrate the same underlying GPT model using OpenAI’s API. Look into LangChain or LlamaIndex as ways to build out a proof-of-concept connecting GPT to your own content
Thanks @novaphil , will do!
My biggest fear is the cost of training with our specific knowledge base.
There’s no fine-tuning (training) costs with this method. Generally that method is not recommended for “educating” GPT on your content.
Instead this uses embeddings/vector databases to lookup related content and inject that in the GPT prompt along with your users prompt. You pay the standard per request pricing
I am assuming what you want is to make the search appear more natural to the user like how people use ChatGPT or Bing. If your data is dynamic (e.g. products changes frequently, new offers depending on season, etc.) I don’t think fine-tuning will help in that case. From what I understand, fine-tuning is more useful if you have static data (e.g. books, manuals, etc.). In your use case, to integrate ChatGPT in the search function, when your user submits an inquiry, using the Text Completion API, it will be transformed into a command for your own search API. Then submit the result + original user inquiry + a prompt to Chat API for the final reply.