Gpt-3 fine-tuning to inject our company's information

Prompting
1

What I want to do is to make a chatbot with gpt-3 davinci in order for our customers to get help through the chatbot instead of calling our c/s center.

As GPT-3 has no information of our own product, a software, I think I need to fine-tune it.

My questions are

  1. How many epochs would it need for getting new information that it hasn’t heard of?

  2. There are many ways to ask something. (ex. Tell me about sth = What is sth)
    Should I include all of possible form in prompt when fine-tuning?
    Or any other suggestions

  3. Is it possible to learn new programming language? (We have our own programming language similar to javasript)

Thank you.

1 Like
2

Embeddings may be the best option to answer questions about any information that you already have stored in a knowledgebase.

This thread answers a lot of common questions about fine-tuning.

3 Likes