Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack

Documenti analoghi

• Large Language Models can Strategically Deceive their Users when Put Under Pressure
  di: Scheurer, Jérémy, et al.
  Pubblicazione: (2023)
• Lessons from Studying Two-Hop Latent Reasoning
  di: Balesni, Mikita, et al.
  Pubblicazione: (2024)
• Reservoir Computing for Fast, Simplified Reinforcement Learning on Memory Tasks
  di: McKee, Kevin
  Pubblicazione: (2024)
• BeHonest: Benchmarking Honesty in Large Language Models
  di: Chern, Steffi, et al.
  Pubblicazione: (2024)
• Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models
  di: Denison, Carson, et al.
  Pubblicazione: (2024)