Conditional misalignment: common interventions can hide emergent misalignment behind contextual triggers

Documenti analoghi

• Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
  di: Betley, Jan, et al.
  Pubblicazione: (2025)
• Tell me about yourself: LLMs are aware of their learned behaviors
  di: Betley, Jan, et al.
  Pubblicazione: (2025)
• Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
  di: Betley, Jan, et al.
  Pubblicazione: (2025)
• Subliminal Learning: Language models transmit behavioral traits via hidden signals in data
  di: Cloud, Alex, et al.
  Pubblicazione: (2025)
• School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMs
  di: Taylor, Mia, et al.
  Pubblicazione: (2025)