building off previous chat | OpenAI | Page 1

ember pewter Apr 1, 2023, 7:51 PM

#

Hello All.
Working with API (both 3.5-turbo and 4) and I'm trying to figure out how to "continue" a chat. Any suggestions would be welcomed.

Example:

When I'm at chat.openai.com/chat
What's the difference between pencil lead classifications such as H and B
And I get back a multi-paragraph answer. I then ask What are the ranges for each? And Chat responds with an on-target mult-paragraph answer (9H - 9B)

Using the API, the same initial question results in essentially the same response. However, the follow-up question is "dumb" as Chat doesn't remember what it just answered.

"That question is too vague" -- 3.5-turbo
"Here are the ranges: Healthy Weight: {blurb}, Underweight: A person is considered underweight if their BMI is below 18.5"; Overweight: {blurb}.

Instead, if I ask "What are the ranges of pencil lead" I get an acceptable answer.

There is an optional param, but it's not for Chat to "keep track of what it was discussing"

user string Optional
A unique identifier representing your end-user, which can help OpenAI to monitor and detect abuse.

trim mesa Apr 2, 2023, 1:56 AM

#

You need add each new message in messages array

fierce citrus Apr 2, 2023, 3:37 AM

#

ember pewter Hello All. Working with API (both 3.5-turbo and 4) and I'm trying to figure out...

You need to keep track of the messages and include them in the next request.
So if your first user message is My name is Paul., and the server responds with Hi Paul., you need to include those two messages in the next request with your new user message. IE. if your next user message is Who are you?
Your request should include
User: My name is Paul.
Assistant: Hi Paul.
User: Who are you?

ember pewter Apr 2, 2023, 4:22 PM

#

Great. Glad that's an option. I wonder if it will cost more tokens?

Ok, so do I need to include the previous questions and answers? And what would the request look like?

#building off previous chat