Saved in:
Bibliographic Details
Main Authors: Barrow, Joe, Patel, Raj, Kharkovski, Misha, Davies, Ben, Schmitt, Ryan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.00276
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915527543226368
author Barrow, Joe
Patel, Raj
Kharkovski, Misha
Davies, Ben
Schmitt, Ryan
author_facet Barrow, Joe
Patel, Raj
Kharkovski, Misha
Davies, Ben
Schmitt, Ryan
contents Black box large language models (LLMs) make information extraction (IE) easy to configure, but hard to trust. Unlike traditional information extraction pipelines, the information "extracted" is not guaranteed to be grounded in the document. To prevent this, this paper introduces the notion of a "safe passage": context generated by the LLM that is both grounded in the document and consistent with the extracted information. This is operationalized via a three-step pipeline, SafePassage, which consists of: (1) an LLM extractor that generates structured entities and their contexts from a document, (2) a string-based global aligner, and (3) a scoring model. Results show that using these three parts in conjunction reduces hallucinations by up to 85% on information extraction tasks with minimal risk of flagging non-hallucinations. High agreement between the SafePassage pipeline and human judgments of extraction quality mean that the pipeline can be dually used to evaluate LLMs. Surprisingly, results also show that using a transformer encoder fine-tuned on a small number of task-specific examples can outperform an LLM scoring model at flagging unsafe passages. These annotations can be collected in as little as 1-2 hours.
format Preprint
id arxiv_https___arxiv_org_abs_2510_00276
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SafePassage: High-Fidelity Information Extraction with Black Box LLMs
Barrow, Joe
Patel, Raj
Kharkovski, Misha
Davies, Ben
Schmitt, Ryan
Computation and Language
Machine Learning
Black box large language models (LLMs) make information extraction (IE) easy to configure, but hard to trust. Unlike traditional information extraction pipelines, the information "extracted" is not guaranteed to be grounded in the document. To prevent this, this paper introduces the notion of a "safe passage": context generated by the LLM that is both grounded in the document and consistent with the extracted information. This is operationalized via a three-step pipeline, SafePassage, which consists of: (1) an LLM extractor that generates structured entities and their contexts from a document, (2) a string-based global aligner, and (3) a scoring model. Results show that using these three parts in conjunction reduces hallucinations by up to 85% on information extraction tasks with minimal risk of flagging non-hallucinations. High agreement between the SafePassage pipeline and human judgments of extraction quality mean that the pipeline can be dually used to evaluate LLMs. Surprisingly, results also show that using a transformer encoder fine-tuned on a small number of task-specific examples can outperform an LLM scoring model at flagging unsafe passages. These annotations can be collected in as little as 1-2 hours.
title SafePassage: High-Fidelity Information Extraction with Black Box LLMs
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2510.00276