Not All Claims Are Equally Risky: FACTOR for Adaptive Verification in Factual Long-Form Generation

The article introduces FACTOR (FACTuality-Oriented Risk-aware Verification), an inference-time model designed to enhance the factual accuracy of long-form text generated by Large Language Models (LLMs) by adapting verification processes based on claim-level uncertainty. FACTOR employs uncertainty estimation, adaptive language inference verification, and candidate re-ranking, demonstrating improved factuality and reduced verification costs on the FactScore benchmark. This approach is significant for practitioners as it offers a model-agnostic solution to optimize verification efforts in LLM outputs, addressing the common issue of unsupported factual claims in generated text.