Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2503.14477 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910916733304832 |
|---|---|
| author | Ji, Ziwei Yu, Lei Koishekenov, Yeskendir Bang, Yejin Hartshorn, Anthony Schelten, Alan Zhang, Cheng Fung, Pascale Cancedda, Nicola |
| author_facet | Ji, Ziwei Yu, Lei Koishekenov, Yeskendir Bang, Yejin Hartshorn, Anthony Schelten, Alan Zhang, Cheng Fung, Pascale Cancedda, Nicola |
| contents | LLMs often adopt an assertive language style also when making false claims. Such ``overconfident hallucinations'' mislead users and erode trust. Achieving the ability to express in language the actual degree of uncertainty around a claim is therefore of great importance. We find that ``verbal uncertainty'' is governed by a single linear feature in the representation space of LLMs, and show that this has only moderate correlation with the actual ``semantic uncertainty'' of the model. We apply this insight and show that (1) the mismatch between semantic and verbal uncertainty is a better predictor of hallucinations than semantic uncertainty alone and (2) we can intervene on verbal uncertainty at inference time and reduce confident hallucinations on short-form answers, achieving an average relative reduction of ~30%. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_14477 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations Ji, Ziwei Yu, Lei Koishekenov, Yeskendir Bang, Yejin Hartshorn, Anthony Schelten, Alan Zhang, Cheng Fung, Pascale Cancedda, Nicola Computation and Language LLMs often adopt an assertive language style also when making false claims. Such ``overconfident hallucinations'' mislead users and erode trust. Achieving the ability to express in language the actual degree of uncertainty around a claim is therefore of great importance. We find that ``verbal uncertainty'' is governed by a single linear feature in the representation space of LLMs, and show that this has only moderate correlation with the actual ``semantic uncertainty'' of the model. We apply this insight and show that (1) the mismatch between semantic and verbal uncertainty is a better predictor of hallucinations than semantic uncertainty alone and (2) we can intervene on verbal uncertainty at inference time and reduce confident hallucinations on short-form answers, achieving an average relative reduction of ~30%. |
| title | Calibrating Verbal Uncertainty as a Linear Feature to Reduce Hallucinations |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2503.14477 |