On Early Detection of Hallucinations in Factual Question Answering

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Snyder, Ben, Moisescu, Marius, Zafar, Muhammad Bilal
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913476187783168
author Snyder, Ben
Moisescu, Marius
Zafar, Muhammad Bilal
author_facet Snyder, Ben
Moisescu, Marius
Zafar, Muhammad Bilal
contents While large language models (LLMs) have taken great strides towards helping humans with a plethora of tasks, hallucinations remain a major impediment towards gaining user trust. The fluency and coherence of model generations even when hallucinating makes detection a difficult task. In this work, we explore if the artifacts associated with the model generations can provide hints that the generation will contain hallucinations. Specifically, we probe LLMs at 1) the inputs via Integrated Gradients based token attribution, 2) the outputs via the Softmax probabilities, and 3) the internal state via self-attention and fully-connected layer activations for signs of hallucinations on open-ended question answering tasks. Our results show that the distributions of these artifacts tend to differ between hallucinated and non-hallucinated generations. Building on this insight, we train binary classifiers that use these artifacts as input features to classify model generations into hallucinations and non-hallucinations. These hallucination classifiers achieve up to $0.80$ AUROC. We also show that tokens preceding a hallucination can already predict the subsequent hallucination even before it occurs.
format Preprint
id arxiv_https___arxiv_org_abs_2312_14183
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle On Early Detection of Hallucinations in Factual Question Answering
Snyder, Ben
Moisescu, Marius
Zafar, Muhammad Bilal
Computation and Language
Artificial Intelligence
While large language models (LLMs) have taken great strides towards helping humans with a plethora of tasks, hallucinations remain a major impediment towards gaining user trust. The fluency and coherence of model generations even when hallucinating makes detection a difficult task. In this work, we explore if the artifacts associated with the model generations can provide hints that the generation will contain hallucinations. Specifically, we probe LLMs at 1) the inputs via Integrated Gradients based token attribution, 2) the outputs via the Softmax probabilities, and 3) the internal state via self-attention and fully-connected layer activations for signs of hallucinations on open-ended question answering tasks. Our results show that the distributions of these artifacts tend to differ between hallucinated and non-hallucinated generations. Building on this insight, we train binary classifiers that use these artifacts as input features to classify model generations into hallucinations and non-hallucinations. These hallucination classifiers achieve up to $0.80$ AUROC. We also show that tokens preceding a hallucination can already predict the subsequent hallucination even before it occurs.
title On Early Detection of Hallucinations in Factual Question Answering
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2312.14183