Comment by flemhans

So I would want to include a big corpus like GPT-3 or this newfangled "Neo" thing but still have it trained to respond to our own customers based on 200k email passages.

How to create a hybrid?