representation-steering

The paper proposes TARGRS, a test-time detoxification framework that learns token-level, transition-aware reward guidance over toxicity trajectories in LLM latent space, achieving up to 69.38% toxicity reduction on RealToxicityPrompts and extending to vision-language models with minimal inference overhead.

This paper demonstrates that Whisper's hallucination failures on silence, noise, or music can be detected and mitigated purely from internal activations using sparse autoencoders, achieving large reductions in hallucination rate without fine-tuning.

This paper introduces Contextual Belief Management (CBM) for LLMs to handle long-term information, proposes the BeliefTrack benchmark for evaluation, and demonstrates that reinforcement learning and representation-level steering significantly reduce belief management failures.