I notice that for embedding, text-embedding-ada-002 is recommended. Yet when it comes to completion, this model underperforms text-davinci-003. Thus my question:
is it a standard practice to vectorize with one model and use another model for completion?
My guess yes but I want to hear your perspectives.