Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAR:Multi-Agent Reflexion Improves Reasoning Abilities in LLMs

Onat Ozer, Grace Wu, Yuchen Wang, Daniel Dosti, Honghao Zhang, Vivi De La Rue

TL;DR

MAR introduces a multi-agent debate framework to overcome degeneration-of-thought in reflexion-based reasoning by substituting a single self-critique loop with diverse, persona-driven critics. Across HotPotQA and HumanEval, MAR achieves notable gains over both baselines and single-agent Reflexion, attaining 47% EM on HotPotQA and 82.6% pass@1 on HumanEval, indicating stronger, more reliable corrections through aggregated, multi-perspective reflections. The work demonstrates that structured disagreement and memory-based aggregation can significantly improve self-improvement for LLMs, albeit at increased computational cost, and discusses societal implications and future directions for scalable, self-refining agents.

Abstract

LLMs have shown the capacity to improve their performance on reasoning tasks through reflecting on their mistakes, and acting with these reflections in mind. However, continual reflections of the same LLM onto itself exhibit degeneration of thought, where the LLM continues to repeat the same errors again and again even with the knowledge that its wrong. To address this problem, we instead introduce multi-agent with multi-persona debators as the method to generate reflections. Through out extensive experimentation, we've found that the leads to better diversity of in the reflections generated by the llm agent. We demonstrate an accuracy of 47% EM HotPot QA (question answering) and 82.7% on HumanEval (programming), both performances surpassing reflection with a single llm.

MAR:Multi-Agent Reflexion Improves Reasoning Abilities in LLMs

TL;DR

MAR introduces a multi-agent debate framework to overcome degeneration-of-thought in reflexion-based reasoning by substituting a single self-critique loop with diverse, persona-driven critics. Across HotPotQA and HumanEval, MAR achieves notable gains over both baselines and single-agent Reflexion, attaining 47% EM on HotPotQA and 82.6% pass@1 on HumanEval, indicating stronger, more reliable corrections through aggregated, multi-perspective reflections. The work demonstrates that structured disagreement and memory-based aggregation can significantly improve self-improvement for LLMs, albeit at increased computational cost, and discusses societal implications and future directions for scalable, self-refining agents.

Abstract

LLMs have shown the capacity to improve their performance on reasoning tasks through reflecting on their mistakes, and acting with these reflections in mind. However, continual reflections of the same LLM onto itself exhibit degeneration of thought, where the LLM continues to repeat the same errors again and again even with the knowledge that its wrong. To address this problem, we instead introduce multi-agent with multi-persona debators as the method to generate reflections. Through out extensive experimentation, we've found that the leads to better diversity of in the reflections generated by the llm agent. We demonstrate an accuracy of 47% EM HotPot QA (question answering) and 82.7% on HumanEval (programming), both performances surpassing reflection with a single llm.
Paper Structure (65 sections, 3 figures, 4 tables)

This paper contains 65 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. Reflexion Framework
  5. Datasets
  6. Performance Metrics
  7. Methodology, Results, and Discussion
  8. Methodology
  9. HotPotQA Setup.
  10. HumanEval Setup.
  11. Results
  12. HotPotQA Replication
  13. HumanEval Replication
  14. Discussion
  15. HotPotQA Discussion
  16. ...and 50 more sections

Figures (3)

  • Figure 1: The Reflexion Architecture. Actor performs an action on the environment. The evaluator gives feedback to reflector LLM, with feedback stored in short-term memory. A new action is then performed starting a new iteration.
  • Figure 2: The Multi-Agent Reflexion (MAR) Architecture. This high-level diagram illustrates the extension of the single-agent Reflexion framework.
  • Figure 3: Comparison of HotPotQA performance across trials for ReAct, Reflexion, and Multi-Agent Reflexion (MAR). The baseline gpt-3.5-Turbo (grey) and the Reflexion replication results (blue) are shown. MAR (red) offers the highest EM improvement.