Decide whether to trust AI memory or search results

Original: TrustMargin: Training-Free Arbitration between Parametric Memory and Retrieved Evidence in Large Language Models

Source: arxiv.org ↗