Domain knowledge and math

From what I understand, to use domain knowledge we would use fine tuning. I want to create a domain knowledge Q&A.

I have a few questions:

  1. Are the dataset I feed to private or is it shared with everyone?
  2. How do we train math? Let’s say demonstrating how to calculate some optimization using matrix.
    a) How do you input into prompt and completion with math?
    b) How do you show greeks mathematically?
  3. Suppose there are some answers the default chatGPT gets it correct, but there are some that is off. In order to reduce cost, I suppose we take the incorrect ones. Is this the right approach?
  4. Is embeddings not available for the free credits?

Thank you.

1 Like
  1. I assume it is private. You’re the only one who have access to the fine-tuned version.
  2. Good question. I think GPT-3 understands LaTeX notation.
  3. I didn’t really understand this question
  4. Embeddings are included in the free credits.
1 Like

Thank you for the response.

Interesting enough, I just asked gpt itself those questions and help me find the best way to reduce cost :joy:

1 Like

What was chatgpt’s response?