Only Say What You Know: Calibration-Aware Generation for Long-Form Factuality

TL;DR

This paper introduces a new paradigm for long-form generation called Exploration-Commitment Decoupling, enabling models to explore information with awareness and generate more factual, reliable content.

Contribution

It proposes the Calibration-Aware Generation framework that decouples exploration from final commitment, improving factuality and efficiency in long-form generation.

Findings

CAG improves factuality by up to 13% across benchmarks.

CAG reduces decoding time by up to 37%.

Decoupling exploration and commitment enhances reliability in generation.

Abstract

Large Reasoning Models achieve strong performance on complex tasks but remain prone to hallucinations, particularly in long-form generation where errors compound across reasoning steps. Existing approaches to improving factuality, including abstention and factuality-driven optimization, follow a \emph{coupled exploration-commitment} paradigm, in which intermediate reasoning is unconditionally propagated to the final output, limiting fine-grained control over information selection and integration. In this paper, we propose an \textbf{Exploration-Commitment Decoupling} paradigm that disentangles knowledge exploration from final commitment, enabling models to explore with awareness while answering cautiously. We instantiate the paradigm with \textbf{Calibration-Aware Generation (CAG)}, a framework that equips models with end-to-end, calibration-aware generation capabilities, by augmenting intermediate reasoning with calibrated reliability estimates and prioritizing reliable content in final outputs. Across five long-form factuality benchmarks and multiple model families, CAG improves factuality by up to 13%, while reducing decoding time by up to 37%. Overall, our work highlights decoupling as a principled approach for more reliable long-form generation, offering directions for trustworthy and self-aware generative systems.