Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features

Documenti analoghi

• CorrSteer: Generation-Time LLM Steering via Correlated Sparse Autoencoder Features
  di: Cho, Seonglae, et al.
  Pubblicazione: (2025)
• Are LLM Uncertainty and Correctness Encoded by the Same Features? A Functional Dissociation via Sparse Autoencoders
  di: Patel, Het, et al.
  Pubblicazione: (2026)
• Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features
  di: McCann, Jordan F.
  Pubblicazione: (2026)
• MCP: A Control-Theoretic Orchestration Framework for Synergistic Efficiency and Interpretability in Multimodal Large Language Models
  di: Zhang, Luyan
  Pubblicazione: (2025)
• Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation
  di: Resck, Lucas, et al.
  Pubblicazione: (2026)