Hi there,
I am trying to integrate chatGPT to my own chatbot. Normally we have classifier and by using that classifier we are getting intent of the sentence (and also top 10 intents). We have tried to do this by azure playgraound and we created a prompt where we gave some examples with their intents and also specify that bot should answer within this intents and. also bring top 3 intents. It is working and trails are satisfactory. We can use API to integrate but there are some difficulties. If we dont give intent and examples chatGPT answer how it define intents and not related how i wish. So it must know how i classify intents. Second for one of my project there are 400 intents and 100.000 sentences. But i want to give 400 intent and maybe 800 examples only. When i prepare this prompt it is like 16000 token for every message my user typed. So I wanted to fine-tune gpt model with my 800 sentence and 400 intents that chatgpt can answer within this context. But here microsoft person direct me to this page. I have read also langchain documents. But i couldnt find a way how to reach my goal without fine-tuning and sending just user text to GPT but not examples and intents. Do you have any idea how i can do that with this cognitive stuff.