Q&A system (embeddings) based on Curie

Documentation
1

Hello All,

These are the excepts from OpenAI Documentation about Curie on Q&A:

Curie is also quite good at answering questions and performing Q&A and as a general service chatbot.

I’m looking for a tutorial that has the following in it:

  • Uses the embedding approach for factual Q&A system
  • Uses Curie (as opposed to Davinci in the cookbook example) as the completions model

Thank you for your time.

1 Like
2

Here us a workbook using Davinci and the latest embedding engine ada-002 (considered better than it’s predecessors, for the most part, at embedding, and has a smaller vector size which is a performance advantage)

1 Like
3

I’d recommend using Davinci 003. Curie can work to summarize embeddings, then you can pass the summaries through Davinci so you can use more context per q&a.

4

Thank you @curt.kennedy & @charlieevert for your feedback.

I was able to get the sample Q/A app up and running which is based on embeddings (using text-embedding-ada-002) and text-davinci-003 model for completions.

Now I’m trying to see if curie can do a better job for building a Q/A system as the OpenAI documentation claims this:

Curie is also quite good at answering questions and performing Q&A and as a general service chatbot.

Any thoughts?

2 Likes
5

Go to the lowest model that you are satisfied with, because it is faster and cheaper.

1 Like