# Stay Focused: Problem Drift in Multi-Agent Debate

arXiv: https://arxiv.org/abs/2502.19559 · HTML: https://mlsanity.com/paper/2502.19559/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- Placed in Social simulation and multi-agent debate, one of 16 directions in LLM agents and tool use.
- Builds on 37 of its 78 references that are in the map.
- First posted on arXiv 26 Feb 2025.

## Placed in

- LLM agents and tool use / [Social simulation and multi-agent debate](https://mlsanity.com/agents/social-simulation-and-multi-agent-debate/)

## Builds on (most cited first, 30 of 37)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://mlsanity.com/paper/2305.14325/) (2023)
- [Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization](https://mlsanity.com/paper/1808.08745/) (2018)
- [Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate](https://mlsanity.com/paper/2305.19118/) (2023)
- [ExpeL: LLM Agents Are Experiential Learners](https://mlsanity.com/paper/2308.10144/) (2023)
- [Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies](https://mlsanity.com/paper/2101.02235/) (2021)
- [Aligning AI With Shared Human Values](https://mlsanity.com/paper/2008.02275/) (2020)
- [Program Induction by Rationale Generation : Learning to Solve and Explain Algebraic Word Problems](https://mlsanity.com/paper/1705.04146/) (2017)
- [How to Fine-Tune BERT for Text Classification?](https://mlsanity.com/paper/1905.05583/) (2019)
- [Emergent Tool Use From Multi-Agent Autocurricula](https://mlsanity.com/paper/1909.07528/) (2019)
- [ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs](https://mlsanity.com/paper/2309.13007/) (2023)
- [Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metrics](https://mlsanity.com/paper/2104.13346/) (2021)
- [ToolQA: A Dataset for LLM Question Answering with External Tools](https://mlsanity.com/paper/2306.13304/) (2023)
- [Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration](https://mlsanity.com/paper/2307.05300/) (2023)
- [Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback](https://mlsanity.com/paper/2305.10142/) (2023)
- [ExpertPrompting: Instructing Large Language Models to be Distinguished Experts](https://mlsanity.com/paper/2305.14688/) (2023)
- [Evaluation of OpenAI o1: Opportunities and Challenges of AGI](https://mlsanity.com/paper/2409.18486/) (2024)
- [Meta-Prompting: Enhancing Language Models with Task-Agnostic Scaffolding](https://mlsanity.com/paper/2401.12954/) (2024)
- [Can Large Language Models Really Improve by Self-critiquing Their Own Plans?](https://mlsanity.com/paper/2310.08118/) (2023)
- [GPT-4 Doesn't Know It's Wrong: An Analysis of Iterative Prompting for Reasoning Problems](https://mlsanity.com/paper/2310.12397/) (2023)
- [Exchange-of-Thought: Enhancing Large Language Model Capabilities through Cross-Model Communication](https://mlsanity.com/paper/2312.01823/) (2023)
- [Corex: Pushing the Boundaries of Complex Reasoning through Multi-Model Collaboration](https://mlsanity.com/paper/2310.00280/) (2023)
- [Learning From Failure: Integrating Negative Examples when Fine-tuning Large Language Models as Agents](https://mlsanity.com/paper/2402.11651/) (2024)

## Built on by (most cited first, 23 of 23)

- [The Illusion of Diminishing Returns: Measuring Long Horizon Execution in LLMs](https://mlsanity.com/paper/2509.09677/) (2025)
- [Literature Review Of Multi-Agent Debate For Problem-Solving](https://mlsanity.com/paper/2506.00066/) (2025)
- [MALLM: Multi-Agent Large Language Models Framework](https://mlsanity.com/paper/2509.11656/) (2025)
- [Amplified Vulnerabilities: Structured Jailbreak Attacks on LLM-based Multi-Agent Debate](https://mlsanity.com/paper/2504.16489/) (2025)
- [Tree-of-Reasoning: Towards Complex Medical Diagnosis via Multi-Agent Reasoning with Evidence Tree](https://mlsanity.com/paper/2508.03038/) (2025)
- [SPaRC: A Spatial Pathfinding Reasoning Challenge](https://mlsanity.com/paper/2505.16686/) (2025)
- [TrinityGuard: A Unified Framework for Safeguarding Multi-Agent Systems](https://mlsanity.com/paper/2603.15408/) (2026)
- [Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting](https://mlsanity.com/paper/2607.00038/) (2026)
- [Reallocating Attention Across Layers to Reduce Multimodal Hallucination](https://mlsanity.com/paper/2510.10285/) (2025)
- [Multi-Agent Reasoning Improves Compute Efficiency: Pareto-Optimal Test-Time Scaling](https://mlsanity.com/paper/2605.01566/) (2026)
- [The Reasoning Trap: An Information-Theoretic Bound on Closed-System Multi-Step LLM Reasoning](https://mlsanity.com/paper/2605.01704/) (2026)
- [Misinformation Propagation in Benign Multi-Agent Systems](https://mlsanity.com/paper/2606.16710/) (2026)
- [From Language Models to World-Acting Systems: Progress and Limits of Agentic AI across Digital, Social, Virtual, and Physical Environments](https://mlsanity.com/paper/2609.04894/) (2026)
- [Diagnosing and Mitigating Compounding Failures in Agentic Persuasion via Taxonomic Strategy Retrieval](https://mlsanity.com/paper/2606.24976/) (2026)
- [Scaling LLM-Driven Multi-Agent Systems: Design Principles and Architectural Scalability Analysis](https://mlsanity.com/paper/2607.27942/) (2026)
- [Loop-Back Authority in LLM Agent Teams: A Paired Experiment on Flat and Hierarchical Coordination](https://mlsanity.com/paper/2609.14767/) (2026)
- [The Prompt Engineering Report Distilled: Quick Start Guide for Life Sciences](https://mlsanity.com/paper/2509.11295/) (2025)
- [Internal vs. External: Comparing Deliberation and Evolution for Multi-Agent Constitutional Design](https://mlsanity.com/paper/2605.09128/) (2026)
- [Enhancing Multi-Agent Communication through Attention Steering with Context Relevance](https://mlsanity.com/paper/2605.30136/) (2026)
- [Do LLM Debates Repeat Arguments Differently Across Languages?](https://mlsanity.com/paper/2607.23442/) (2026)
- [Where Reasoning Diverges: Localized Multi-Agent Debate for Multi-Hop Question Answering](https://mlsanity.com/paper/2608.01463/) (2026)
- [What Does Multi-Agent LLM Debate Actually Change? A Layered Analysis of Disagreement and Answer Quality](https://mlsanity.com/paper/2609.08016/) (2026)
- [From Process Loss to Assembly Bonus: Human-Grounded Diagnosis of Multi-Agent LLM Collaboration](https://mlsanity.com/paper/2609.13261/) (2026)

## Abstract (arXiv)

Multi-agent debate - multiple instances of large language models discussing problems in turn-based interaction - has shown promise for solving knowledge and reasoning tasks. However, these methods show limitations when solving complex problems that require longer reasoning chains. We analyze how multi-agent debate drifts away from the initial problem over multiple turns, thus harming task performance. We define this phenomenon as problem drift and quantify its presence across ten tasks (i.e., three generative, three knowledge, three reasoning, and one instruction-following task). We find that generative tasks drift often due to the subjectivity of the answer space (76-89%), compared to high-complexity tasks (7-21%). To identify the reasons, eight human experts analyze 170 multi-agent debates suffering from problem drift. We find the most common issues related to this drift are the lack of progress (35% of cases), low-quality feedback (26% of cases), and a lack of clarity (25% of cases). We propose DRIFTJudge, an LLM-as-a-judge method, as a first baseline to detect problem drift. We also propose DRIFTPolicy, which mitigates 31% of problem drift cases. Our study is a step toward understanding a key limitation of multi-agent debate, highlighting why longer debates can harm task performance and how problem drift could be addressed.