FaithBench: A Diverse Hallucination Benchmark for Summarization by Modern LLMs

TL;DR

FaithBench is a new benchmark for evaluating hallucinations in summarization by modern LLMs, highlighting the diversity of models and the challenges in detection accuracy.

Contribution

Introduces FaithBench, a diverse and challenging hallucination benchmark with human annotations, covering 10 LLMs from 8 families for improved evaluation.

Findings

GPT-4o and GPT-3.5-Turbo produce fewer hallucinations.

Current detection models have about 50% accuracy on FaithBench.

Significant room for improvement in hallucination detection.

Abstract

Summarization is one of the most common tasks performed by large language models (LLMs), especially in applications like Retrieval-Augmented Generation (RAG). However, existing evaluations of hallucinations in LLM-generated summaries, and evaluations of hallucination detection models both suffer from a lack of diversity and recency in the LLM and LLM families considered. This paper introduces FaithBench, a summarization hallucination benchmark comprising challenging hallucinations made by 10 modern LLMs from 8 different families, with ground truth annotations by human experts. ``Challenging'' here means summaries on which popular, state-of-the-art hallucination detection models, including GPT-4o-as-a-judge, disagreed on. Our results show GPT-4o and GPT-3.5-Turbo produce the least hallucinations. However, even the best hallucination detection models have near 50\% accuracies on FaithBench, indicating lots of room for future improvement. The repo is https://github.com/vectara/FaithBench