From Out-of-Distribution Detection to Hallucination Detection: A Geometric View
Litian Liu, Reza Pourreza, Yubing Jian, Yao Qin, Roland Memisevic · 2026
Read From Out-of-Distribution Detection to Hallucination Detection: A Geometric View by Litian Liu, Reza Pourreza, Yubing Jian, Yao Qin, Roland Memisevic (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-2602-07253.txt.
Description
Hallucination detectors work well for question answering but falter on thinking tasks, where errors hide in chains of thought. Litian Liu and four co-authors reframe the problem through out-of-spread detection, treating next-token prediction as classification and adapting vision-style OOD techniques to language models.
The resulting detectors need no training and judge single samples, yet they achieve strong accuracy on thinking hallucinations. The reframing itself is the contribution: a well-studied detection problem, properly adapted, covers a newer safety problem. Teams building trustworthy language models get a lightweight detector and a fresh angle on a stubborn failure mode.
Catalog record
| CREATOR | Litian Liu, Reza Pourreza, Yubing Jian, Yao Qin, Roland Memisevic |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07253