How can I customize the gpt

I’m developing a chatbot for my own organization using the ChatGPT API. I have two questions for you: 1- How can I customize the gpt for my own website, I want it to generate the answers it gives according to the content of my website.
2- When the user requests a suggestion, how can I make a suggestion in the categories on my site?

(I am using ChatGPT API gpt-3.5-turbo model.)

Use Embeddings.

I am using ChatGPT API gpt-3.5-turbo model I guess there is no embedding for this model.

You should start out with the tutorial settings:

COMPLETIONS_MODEL = "text-davinci-003"
EMBEDDING_MODEL = "text-embedding-ada-002"

Embedding models create vectors, they are independent of the underlying GPT.

1 Like

meaning, embedding
and i can use it with GPT3.5-Turbo?

Yeah, you can use embeddings with turbo. What the embeddings do is create an object (a vector) on each chunk of text you give it. You then send the incoming request to the embedding engine, get another object (a vector) and then correlate this with all the other vectors. Then you use the text behind the top vectors (most related), feed these into turbo, or other GPT instance, in the prompt and ask turbo or GPT to respond with this information as the context. So now turbo has knowledge of your system and facts, instead of only using what it was trained on the internet with.

1 Like

How can I use embeddings with Chat GPT 3-5 Turbo
{role: “user”, content: “Context:<embedding results>. \nQuestion: <user’s last question>”}

The first way that came to my mind was something like this. I’m not really sure if that’s true.

Did you see the tutorial link above? It showed how to extend GPT-3 to current events it wasn’t trained on. So put similar things in the 3.5-turbo prompt, you will have to experiment on which one to get the best results. There are various posts on this forum on how to use turbo like people use davinci.

If I were you though, I wouldn’t try to learn this with turbo. Learn it with davinci and then see if you can switch over to turbo (because it’s cheaper). But we know davinci works.

thanks. i think there is no standard way to do this described by the official (OpenAI). I will try to apply it as shown above for now

1 Like

I have one more question I want to ask you. How to distinguish: Will a query be answered according to Embeddings or will ChatGPT default.