I think it's the feedback from RLHF that is mostly responsible, but it only works if the base model is large. Never seen a small model doing good conversation.

I think it's the feedback from RLHF that is mostly responsible, but it only works if the base model is large. Never seen a small model doing good conversation. They can do ok for classification and open book question answering, but generating long form coherent text is hard.