RLHF helps with getting the model in the "mood" to output responses in a certain style that is found to be helpful by users. i.e how to write a poem, or email.

RLHF helps with getting the model in the "mood" to output responses in a certain style that is found to be helpful by users. i.e how to write a poem, or email. It doesn't increase it's knowledge of the world or increase it's capabilities.