Forward-Looking Active Retrieval: Reimagining Uncertainty as an Epistemic Signal in Language Models

Large language models, despite their impressive capacity for language understanding and generation, frequently hallucinate factual content. While retrieval augmented generation has emerged as a primary mitigation strategy, most implementations rely on a static retrieve then generate paradigm. This approach retrieves relevant documents once at the outset and generates the complete response conditioning on this fixed context. Such methods suffice for short form tasks like factoid question answering, where information needs are explicit in the query. However, they falter