Why ChatGPT Feels Helpful Instead of Just Smart

SFT teaches a model to follow instructions. RLHF teaches it to understand what humans actually want and those two things are not the same… Continue reading on Medium »