MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866909819719385088 |
|---|---|
| author | Liu, Gabrielle Kaili-May Yona, Gal Caciularu, Avi Szpektor, Idan Rudner, Tim G. J. Cohan, Arman |
| author_facet | Liu, Gabrielle Kaili-May Yona, Gal Caciularu, Avi Szpektor, Idan Rudner, Tim G. J. Cohan, Arman |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_24858 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs Liu, Gabrielle Kaili-May Yona, Gal Caciularu, Avi Szpektor, Idan Rudner, Tim G. J. Cohan, Arman Computation and Language Machine Learning 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. |
| title | MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2505.24858 |