# ranking

DPO & Direct Alignment Attacks

Direct Preference Optimization vulnerabilities, how DPO differs from RLHF in attack surface, preference pair poisoning, and ranking manipulation techniques.