LLMs Should Express Uncertainty Explicitly

2026-04-06

AI Summary

This work addresses the challenge that large language models often fail to explicitly convey their confidence when making decisions involving uncertainty—such as abstaining, retrieving external information, or verifying outputs. The paper introduces a novel approach that formulates uncertainty as a task-adaptive, controllable communication interface, combining complementary global and local mechanisms. The global mechanism calibrates output confidence to assess overall answer reliability, while the local mechanism inserts explicit <uncertain> tokens during inference to pinpoint precise intervention points. By integrating verbalized confidence, explicit uncertainty markers, adaptive retrieval-augmented generation (RAG) control, and internal representation analysis, the method significantly improves model calibration, reduces overconfident errors, enhances error coverage, and enables retrieval triggering with both high selectivity and high recall, thereby optimizing the reliability of final decisions.

Abstract

Large language models are increasingly used in settings where uncertainty must drive decisions such as abstention, retrieval, and verification. Most existing methods treat uncertainty as a latent quantity to estimate after generation rather than a signal the model is trained to express. We instead study uncertainty as an interface for control. We compare two complementary interfaces: a global interface, where the model verbalizes a calibrated confidence score for its final answer, and a local interface, where the model emits an explicit <uncertain> marker during reasoning when it enters a high-risk state. These interfaces provide different but complementary benefits. Verbalized confidence substantially improves calibration, reduces overconfident errors, and yields the strongest overall Adaptive RAG controller while using retrieval more selectively. Reasoning-time uncertainty signaling makes previously silent failures visible during generation, improves wrong-answer coverage, and provides an effective high-recall retrieval trigger. Our findings further show that the two interfaces work differently internally: verbal confidence mainly refines how existing uncertainty is decoded, whereas reasoning-time signaling induces a broader late-layer reorganization. Together, these results suggest that effective uncertainty in LLMs should be trained as task-matched communication: global confidence for deciding whether to trust a final answer, and local signals for deciding when intervention is needed.

Problem

• uncertainty expression
• large language models
• decision-driven uncertainty
• calibrated confidence
• reasoning-time signaling

Innovation

• uncertainty expression
• confidence calibration
• reasoning-time signaling
• Adaptive RAG
• LLM control interface