RLHF (reinforcement learning from human feedback) is used to:

ASpeed up graphics cards BAlign model behaviour with human preferences CCompress training data DEncrypt conversations