ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models

TL;DR

This paper presents an iterative self-training framework for scalable and accurate hallucination annotation in large language models, significantly improving detection and mitigation of hallucinations across domains.

Contribution

It introduces a novel EM-based self-training approach that scales hallucination annotation datasets and enhances annotator accuracy, surpassing GPT-4 in detection performance.

Findings

The hallucination annotator outperforms GPT-4 in detection accuracy.

The framework achieves state-of-the-art results on HaluEval and HalluQA.

Mitigates LLM hallucinations, improving NLI scores from 25% to 37%.

Abstract

Large language models (LLMs) exhibit hallucinations in long-form question-answering tasks across various domains and wide applications. Current hallucination detection and mitigation datasets are limited in domains and sizes, which struggle to scale due to prohibitive labor costs and insufficient reliability of existing hallucination annotators. To facilitate the scalable oversight of LLM hallucinations, this paper introduces an iterative self-training framework that simultaneously and progressively scales up the hallucination annotation dataset and improves the accuracy of the hallucination annotator. Based on the Expectation Maximization (EM) algorithm, in each iteration, the framework first applies a hallucination annotation pipeline to annotate a scaled dataset and then trains a more accurate hallucination annotator on the dataset. This new hallucination annotator is adopted in the hallucination annotation pipeline used for the next iteration. Extensive experimental results demonstrate that the finally obtained hallucination annotator with only 7B parameters surpasses the performance of GPT-4 and obtains new state-of-the-art hallucination detection results on HaluEval and HalluQA by zero-shot inference. Such an annotator can not only evaluate the hallucination levels of various LLMs on the large-scale dataset but also help to mitigate the hallucination of LLMs generations, with the Natural Language Inference (NLI) metric increasing from 25% to 37% on HaluEval.