Trust in One Round: Confidence Estimation for Large Language Models via Structural Signals Conference: Proceedings of the ACM Web Conference 2026; April 13–17, 2026; Dubai, United Arab Emirates Proceedings of the ACM Web Conference 2026 (WWW ’26), April 13–17, 2026, Dubai, United Arab Emirates DOI: 10.1145/3774904.3793055 ISBN: 979-8-4007-2307-0/2026/04 CCS: Computing methodologies Neural networks CCS: Computing methodologies Knowledge representation and reasoning CCS: Computing methodologies Learning latent representations
Pengyue Yang, Jiawen Wen, Haolin Jin, Linghan Huang, Huaming Chen, Ling Chen · 2026
Read Trust in One Round: Confidence Estimation for Large Language Models via Structural Signals Conference: Proceedings of the ACM Web Conference 2026; April 13–17, 2026; Dubai, United Arab Emirates Proceedings of the ACM Web Conference 2026 (WWW ’26), April 13–17, 2026, Dubai, United Arab Emirates DOI: 10.1145/3774904.3793055 ISBN: 979-8-4007-2307-0/2026/04 CCS: Computing methodologies Neural networks CCS: Computing methodologies Knowledge representation and reasoning CCS: Computing methodologies Learning latent representations by Pengyue Yang, Jiawen Wen, Haolin Jin, Linghan Huang, Huaming Chen, Ling Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00977.txt.
Description
Pengyue Yang and colleagues judge language model answers from the shape of activity, not from token odds alone. Their Structural Confidence frame reads mixed signals from a model's final-layer hidden-state path, blending spectral, local-variation, and shape traits. These traits catch stability patterns that odds and sentence embeddings miss. Tests across four benchmarks, FEVER, SciFact, WikiBio-hallucination, and TruthfulQA, show strong results against baselines on AUROC and AUPR. Unlike sampling methods that need many runs plus a helper model, this frame needs one deterministic forward pass.
Catalog record
| CREATOR | Pengyue Yang, Jiawen Wen, Haolin Jin, Linghan Huang, Huaming Chen, Ling Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 53133 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.00977