Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level Diagnosis

Documenti analoghi

• Token-Level Policy Optimization: Linking Group-Level Rewards to Token-Level Aggregation via Sequence-Level Likelihood
  di: Lin, Xingyu, et al.
  Pubblicazione: (2026)
• Token-Level Policy Optimization: Linking Group-Level Rewards to Token-Level Aggregation via Markov Likelihood
  di: Lin, Xingyu, et al.
  Pubblicazione: (2025)
• HINTBench: Horizon-agent Intrinsic Non-attack Trajectory Benchmark
  di: Wang, Jiacheng, et al.
  Pubblicazione: (2026)
• Resisting Contextual Interference in RAG via Parametric-Knowledge Reinforcement
  di: Lin, Chenyu, et al.
  Pubblicazione: (2025)
• FaithfulRAG: Fact-Level Conflict Modeling for Context-Faithful Retrieval-Augmented Generation
  di: Zhang, Qinggang, et al.
  Pubblicazione: (2025)