Hi, I am building a chatbot using gpt4-turbo.
- user asks question
- answer is returned.
- user may ask follow up question
- answer is modified accordingly.
my chatbot works well in this scenario.
However, when a new question is asked in the same thread, it doesnt work well.
What is the best way to use to identify a shift in topic?
- Improve prompt someway?
- Use non-chatgpt techniques to filter this?
- or something else?
Have you made assistant instructions for the AI specifically telling it that the answer will be modified by the following inputs? If doing so, you’ve kind of sealed your fate of the AI always doing that.
As long as the AI isn’t hung up on what it previously produced (where it can’t improve its writing) it should not need more than a change in the topic for the AI to alter course and disregard the topic of prior inputs (unless user told the AI “permanently act like a princess”).
Because “assistants” endpoint threads grow until a chat can be expensive per turn, I would encourage abandonment with a “clear chat” or “new topic” button in the UI.
ok thank you!
IS it okay to change the prompt in each call for the same thread?
Like if a person asks about dance, I have certain set of instruction that I need to use.
But if a person asks about sports, I have different instructions.
In one thread, is it wrong to change the context?
That can be innovative, either adding additional_instructions, or just placing instructions that replace.
However, something intelligent outside of the assistant has to decide what the user input and the current conversation path dictate as far as “response mode”.