arXiv AI recent: Navigating Unreliable Parametric and Contextual Knowledge: Explicit Knowledge Conflict Resolution for LLM Inference
The authors propose a new framework called MACR for resolving knowledge conflicts in large language models. MACR combines an adaptive knowledge assessment using a modified semantic entrop...
Large language models (LLMs) rely on extensive parametric knowledge and in‑context learning, but integrating external information can create conflicts between internal knowledge and external contexts, as well as among multiple external contexts. Existing methods typically assume either the model...