Generative RLHF-V: Learning Principles from Multi-modal Human Preference | Read Paper on Bytez