I would like to start a discussion around what use cases you would use InstructGPT model instead of GPT 3.5 Turbo and why those uses cases are more suited for InstructGPT. I’ve looked over a bunch of documentation about InstructGPT, but I’m still not understanding when to use it.
The instruct model is much more terse and concise, it just does what it is instructed to do.
gpt-3.5-turbo is chatty, it will talk to you beyond what is asked.
So, if you want a model to do something the instruct model will typically work better and use fewer tokens doing so.
If you want a model to interact with users in a natural, conversational manner the chat model will be much better.