<ul><li>Representation Engineering (RepE) is a powerful paradigm for enhancing AI transparency.</li><li>In Vision-Language Models (VLMs), RepE can address challenges related to visual input overriding factual linguistic knowledge.</li><li>Theoretical framework explaining stability of neural activity in VLMs is developed using principal eigenvector.</li><li>This work transforms RepE into a structured theoretical framework, opening new directions for improving AI robustness, fairness, and transparency.</li></ul>

Why Representation Engineering Works: A Theoretical and Empirical Study in Vision-Language Models

Discover more