作者位次 #2/5ACL 2026Findings· Findings of ACL
Reducing Hallucinations in LLM s via Factuality-Aware Preference Learning
Sindhuja Chaduvula · Ahmed Y. Radwan · Azib Farooq · Yani Ioannou · shaina raza
摘要
Preference alignment methods such as RLHF and Direct Preference Optimization (DPO) improve instruction following, but they can also reinforce hallucinations when preference judgments reward fluency and confidence over factual correctness. We introduce F-DPO (Factuality-aware Direct Preference Opt...