Small Agent Can Also Rock! Empowering Small Language Models as Hallucination Detector
This addresses the problem of hallucination detection for users relying on smaller, open-source LLMs, offering a competitive alternative to powerful closed-source models like GPT-4, though it is incremental as it builds on existing agent and tool-use paradigms.
The paper tackles hallucination detection in language models by proposing HaluAgent, a framework that enables smaller LLMs to actively use tools for detecting multiple hallucination types, achieving performance comparable to or higher than GPT-4 on various tasks and datasets with only 2K tuning samples.
Hallucination detection is a challenging task for large language models (LLMs), and existing studies heavily rely on powerful closed-source LLMs such as GPT-4. In this paper, we propose an autonomous LLM-based agent framework, called HaluAgent, which enables relatively smaller LLMs (e.g. Baichuan2-Chat 7B) to actively select suitable tools for detecting multiple hallucination types such as text, code, and mathematical expression. In HaluAgent, we integrate the LLM, multi-functional toolbox, and design a fine-grained three-stage detection framework along with memory mechanism. To facilitate the effectiveness of HaluAgent, we leverage existing Chinese and English datasets to synthesize detection trajectories for fine-tuning, which endows HaluAgent with the capability for bilingual hallucination detection. Extensive experiments demonstrate that only using 2K samples for tuning LLMs, HaluAgent can perform hallucination detection on various types of tasks and datasets, achieving performance comparable to or even higher than GPT-4 without tool enhancements on both in-domain and out-of-domain datasets. We release our dataset and code at https://github.com/RUCAIBox/HaluAgent.