Probabilistic Concept-Aware Steering for Trustworthy LLM Inference

A new framework for precise and reliable control over LLM behavior through probabilistic activation steering.

Steering vectors (SVs), an inference-time intervention technique for large language models (LLMs), guide the generation process by adding a concept-specific direction vector to intermediate activations during inference. However, existing SV methods frequently yield representation...

ArXiv CS.AI2026-07-22