Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations

Documenti analoghi

• Sowing the Wind, Reaping the Whirlwind: The Impact of Editing Language Models
  di: Hazra, Rima, et al.
  Pubblicazione: (2024)
• Soteria: Language-Specific Functional Parameter Steering for Multilingual Safety Alignment
  di: Banerjee, Somnath, et al.
  Pubblicazione: (2025)
• SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models
  di: Banerjee, Somnath, et al.
  Pubblicazione: (2024)
• ProSocialAlign: Preference Conditioned Test Time Alignment in Language Models
  di: Banerjee, Somnath, et al.
  Pubblicazione: (2025)
• Bridging the Multilingual Safety Divide: Efficient, Culturally-Aware Alignment for Global South Languages
  di: Banerjee, Somnath, et al.
  Pubblicazione: (2026)