Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2412.10246 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915533051396096 |
|---|---|
| author | Kim, Hazel Lamb, Tom A. Bibi, Adel Torr, Philip Gal, Yarin |
| author_facet | Kim, Hazel Lamb, Tom A. Bibi, Adel Torr, Philip Gal, Yarin |
| contents | Large language models (LLMs) frequently generate confident yet inaccurate responses, introducing significant risks for deployment in safety-critical domains. We present a novel, test-time approach to detecting model hallucination through systematic analysis of information flow across model layers. We target cases when LLMs process inputs with ambiguous or insufficient context. Our investigation reveals that hallucination manifests as usable information deficiencies in inter-layer transmissions. While existing approaches primarily focus on final-layer output analysis, we demonstrate that tracking cross-layer information dynamics ($\mathcal{L}$I) provides robust indicators of model reliability, accounting for both information gain and loss during computation. $\mathcal{L}$I integrates easily with pretrained LLMs without requiring additional training or architectural modifications. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_10246 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Detecting LLM Hallucination Through Layer-wise Information Deficiency: Analysis of Ambiguous Prompts and Unanswerable Questions Kim, Hazel Lamb, Tom A. Bibi, Adel Torr, Philip Gal, Yarin Machine Learning Large language models (LLMs) frequently generate confident yet inaccurate responses, introducing significant risks for deployment in safety-critical domains. We present a novel, test-time approach to detecting model hallucination through systematic analysis of information flow across model layers. We target cases when LLMs process inputs with ambiguous or insufficient context. Our investigation reveals that hallucination manifests as usable information deficiencies in inter-layer transmissions. While existing approaches primarily focus on final-layer output analysis, we demonstrate that tracking cross-layer information dynamics ($\mathcal{L}$I) provides robust indicators of model reliability, accounting for both information gain and loss during computation. $\mathcal{L}$I integrates easily with pretrained LLMs without requiring additional training or architectural modifications. |
| title | Detecting LLM Hallucination Through Layer-wise Information Deficiency: Analysis of Ambiguous Prompts and Unanswerable Questions |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2412.10246 |