Too Good to Be Real? Diagnosing and Reducing the Gap Between AI Preference and Real User Engagement
This work proposes Ontology-Masked Reasoning Autoencoding (OMRA), a controlled intervention that masks and reconstructs over-explained spans while preserving stance, factual content, and coherence and reduces the measured gap between AI preference and real user engagement.
Xing-Lang Zhang, Yuan-Meng Xiang, Yun-Yao Zhang et al.
· 0 citations