Adaptive Activation Cancellation for Hallucination Mitigation in Large Language Models
arXiv:2603.10195v1 Announce Type: new Abstract: Large Language Models frequently generate fluent but factually incorrect text. We propose Adaptive Activation Cancellation (AAC), a real-time inference-time framework …
Eric Yocam, Varghese Vaidyan, Gurcan Comert, Paris Kalathas, Yong Wang, Judith L. Mwakalonge
조회수 45회