MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination

TL;DR

MARCH introduces a multi-agent reinforcement learning framework with three specialized agents to reduce hallucinations in LLMs by enforcing factual consistency through deliberate information asymmetry, significantly improving reliability.

Contribution

The paper proposes a novel multi-agent reinforcement learning framework, MARCH, that leverages agent specialization and information asymmetry to effectively reduce hallucinations in LLMs.

Findings

MARCH significantly lowers hallucination rates across benchmarks.

An 8B-parameter LLM with MARCH matches performance of larger closed-source models.

The multi-agent approach enhances factual adherence through co-evolution.

Abstract

Hallucination remains a critical bottleneck for large language models (LLMs), undermining their reliability in real-world applications, especially in Retrieval-Augmented Generation (RAG) systems. While existing hallucination detection methods employ LLM-as-a-judge to verify LLM outputs against retrieved evidence, they suffer from inherent confirmation bias, where the verifier inadvertently reproduces the errors of the original generation. To address this, we introduce Multi-Agent Reinforced Self-Check for Hallucination (MARCH), a framework that enforces rigorous factual alignment by leveraging deliberate information asymmetry. MARCH orchestrates a collaborative pipeline of three specialized agents: a Solver, a Proposer, and a Checker. The Solver generates an initial RAG response, which the Proposer decomposes into claim-level verifiable atomic propositions. Crucially, the Checker validates these propositions against retrieved evidence in isolation, deprived of the Solver's original output. This well-crafted information asymmetry scheme breaks the cycle of self-confirmation bias. By training this pipeline with multi-agent reinforcement learning (MARL), we enable the agents to co-evolve and optimize factual adherence. Extensive experiments across hallucination benchmarks demonstrate that MARCH substantially reduces hallucination rates. Notably, an 8B-parameter LLM equipped with MARCH achieves performance competitive with powerful closed-source models. MARCH paves a scalable path for factual self-improvement of LLMs through co-evolution. The code is at https://github.com/Qwen-Applications/MARCH.