Thanks for the update.
Interesting. so the chatgpt model is meant for the communication side while the others are meant for the background services. Makes sense!
It seems like the ideal paradigm now involves skeleton functions from the communication to invoke these background models if you don’t have any sort of classifier. Less token usage and better accuracy?
It would be nice if we could separate models for different tasks using the assistants framework. For example the “chatty” model is typically used, but when a function task is potentially invoked the more “technical” model can be used in it’s place