MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs
This addresses the critical issue of unreliable uncertainty communication in LLMs for users who rely on them for trustworthy information, representing a novel method rather than an incremental improvement.
The study tackled the problem of LLMs using assertive language for false claims by benchmarking their ability to faithfully express uncertainty, finding that current methods largely fail and can even harm calibration. The result was the introduction of MetaFaith, a prompt-based approach that improved faithfulness by up to 61% and achieved an 83% win rate over original generations.
A critical component in the trustworthiness of LLMs is reliable uncertainty communication, yet LLMs often use assertive language when conveying false claims, leading to over-reliance and eroded trust. We present the first systematic study of $\textit{faithful confidence calibration}$ of LLMs, benchmarking models' ability to use linguistic expressions of uncertainty that $\textit{faithfully reflect}$ their intrinsic uncertainty, across a comprehensive array of models, datasets, and prompting strategies. Our results demonstrate that LLMs largely fail at this task, and that existing interventions are insufficient: standard prompt approaches provide only marginal gains, and existing, factuality-based calibration techniques can even harm faithful calibration. To address this critical gap, we introduce MetaFaith, a novel prompt-based calibration approach inspired by human metacognition. We show that MetaFaith robustly improves faithful calibration across diverse models and task domains, enabling up to 61% improvement in faithfulness and achieving an 83% win rate over original generations as judged by humans.