HalluZig : Hallucination Detection using Zigzag Persistence
Shreyas N. Samaga Gilberto Gonzalez Arroyo Tamal K. Dey · 2026
Read HalluZig : Hallucination Detection using Zigzag Persistence by Shreyas N. Samaga Gilberto Gonzalez Arroyo Tamal K. Dey (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-2601-01552.txt.
Description
Samaga, Gonzalez Arroyo, and Dey detect made-up text from language models by watching how attention changes inside the model. Most checkers read only the final answer, yet errors may start in hidden steps before words appear. The team treats the stack of attention maps across layers as a changing network and studies its shape.
A tool from topology, called zigzag persistence, turns that changing shape into a signature. True and false answers, they argue, leave different signatures. On several test sets, their HalluZig method beat strong rivals. The signature also carried across models, and even part of the network depth gave enough signal to spot false output.
Catalog record
| CREATOR | Shreyas N. Samaga Gilberto Gonzalez Arroyo Tamal K. Dey |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-01552