The paper shows that fine‑tuning a large language model (LLM) with a debate framework—where a generator and a critic compete and a weaker LLM judge adjudicates—reduces reward hacking compared to standard reinforcement learning from AI feedback (RLAIF). In experiments on mathematics tasks, the debate approach keeps the judge’s performance stable, achieving a 45% higher peak validation accuracy than the RLAIF baseline and mitigating the rapid exploitation of judge errors. Additional findings indicate that weakening the judge speeds hacking unless countered by extra debate rounds, that debate can override misalignment prompts, and that word‑limit constraints on critiques help balance the game and prevent judge hacking.
whyItMatters":"The study demonstrates a practical method to curb reward hacking in RL‑based AI systems, addressing a key obstacle for safely scaling AI oversight."
By Zachary Kenton, Lili Janzer, Rory Greig, Tian Huey Teh, Kirill Tyshchuk, Jonah Brown-Cohen, Harri Edwards, Senthooran Rajamanoharan, Noah Y. Siegel, Natasha Jaques, Rohin Shah
MADBench is a benchmark that evaluates the security of Multi-Agent Debate (MAD) systems, which allow large language models to exchange and critique answers. The study categorizes attacks into a layered taxonomy aligned with the MAD workflow and tests six attack families across 356 source tasks and 3,958 test cases. Results indicate that while MAD can reduce answer accuracy attacks compared to single-agent baselines, it may amplify unauthorized reads or writes, and even with collusion among agents, the final answer changes from correct to wrong only 28.30% of the time.
By Yuwan Liu, Jiaming Zhang, Yue Huang, Sisi Duan
The study evaluates multi‑agent debate (MAD) in small language models, testing whether cognitive diversity—via personas, sampling temperature, or model identity—drives performance gains. Across 23 models, five tasks, and over 5,500 runs, MAD consistently outperforms single‑model inference but, when matched for generation budget, it ties or falls behind self‑consistency sampling, with persona prompting actually reducing accuracy. The authors find that MAD’s benefits largely stem from the first answer exchange and that many reported gains are due to ensemble‑sampling effects rather than true diversity, highlighting the need for budget‑matched, contamination‑checked baselines.
whyItMatters":"The findings clarify that MAD’s perceived advantages may be overestimated and that future debate mechanisms must be evaluated against rigorous, budget‑matched baselines to ensure genuine performance improvements."
By Leonardo Ferreira, Gardenia Liu, Kaden Zheng
arXiv:2601. 05746v2 Announce Type: replace Abstract: Recent years have witnessed the rapid development of Large Language Model-based Multi-Agent Systems (MAS), which excel at collaborative decision-making and complex problem-solving.
By Zhenghao Li, Zhi Zheng, Wei Chen, Jielun Zhao, Yong Chen, Tong Xu, Enhong Chen
MABPD (Multi‑Agent Bias Probing & Detection) is a training‑free pipeline that uses three specialized large language model agents to analyze news articles from complementary perspectives and resolve disagreements via a Structured Argument Debate (SAD) protocol. SAD imposes an asymmetric burden of proof—biased claims lacking grounded textual evidence receive zero weight—along with role‑weighted voting and post‑consensus verification, replacing task‑specific supervised decision boundaries. Ablation studies show that the debate module alone accounts for up to a 10.6‑point F1 gain, and on the BABE benchmark MABPD attains 83.4% macro F1, within 0.7 percentage points of the supervised state‑of‑the‑art, while achieving 75.0% zero‑shot accuracy on the SemEval 2019 HyperPartisan corpus.
By Garvit Joshi (Graphic Era University, Dehradun, India), Stavya Dhyani (Graphic Era University, Dehradun, India), Jasmine (Graphic Era University, Dehradun, India), Arun Chauhan (Graphic Era University, Dehradun, India)
arXiv:2510. 20963v2 Announce Type: replace Abstract: Multi-agent debate (MAD) was proposed as a promising approach for ensembling the wisdom of multiple large language models (LLMs) to improve reasoning and provide effective supervision to superhuman LLMs.
By Yongqiang Chen, Gang Niu, James Cheng, Bo Han, Masashi Sugiyama
arXiv:2609.08016v1 Announce Type: new
Abstract: Multi-agent debate, in which several LLMs exchange arguments before answering, is widely assumed to improve answer quality by surfacing genuine disagre...
By Chen Qian
arXiv:2606. 00005v1 Announce Type: new Abstract: We present the Consilium Protocol, a Byzantine Fault Tolerance-derived architecture for structured multi-model AI deliberation that treats inter-model disagreement as epistemic signal rather than error.
By VD Doske
arXiv:2607. 18006v1 Announce Type: cross Abstract: Large language models achieve strong reasoning performance, but often at prohibitive training cost - a challenge that is especially acute for compact models ($\leq 4 \, \mathrm{B}$ parameters) trained under limited budgets.
By Martino M. L. Pulici, Cuong Xuan Chu, Evgeny Kharlamov, Zifeng Ding, Volker Tresp, Yunpu Ma
arXiv:2601. 19921v2 Announce Type: replace-cross Abstract: Multi-agent debate (MAD) is widely used to improve large language model (LLM) performance through test-time scaling, yet recent work shows that vanilla MAD often underperforms simple majority vote despite higher computational cost.
By Xiaochen Zhu, Caiqi Zhang, Yizhou Chi, Tom Stafford, Nigel Collier, Andreas Vlachos
arXiv:2606. 15390v1 Announce Type: cross Abstract: LLM agents can improve without weight updates by accumulating natural-language skills from experience, but current systems entrust every decision about which skills to keep and how to apply them to LLM judgment alone.
By Yixuan Wang, Yiyang Zhou, Yiming Liang, Congyu Zhang, Fuxiao Liu, Jiawei Zhou, Huaxiu Yao
The paper introduces RuVerBench, a benchmark with 2,458 instances for evaluating the reliability of Large Language Models acting as judges (LaaJ) in verifying rubric compliance within agentic scenarios such as deep research and agentic coding. It systematically meta‑evaluates frontier LLMs, revealing that even the most advanced models perform well yet still produce substantial noise. The study also examines how prompt design, batching, and majority voting affect verification accuracy, noting that weaker models are more prompt‑sensitive, batched verification trades accuracy for efficiency, and majority voting offers diminishing returns.
By Yangda Peng, Yunjia Qi, Haotian Xia, Guanzhong He, Xintong Shi, Richeng Xuan, Songyuanyi Lu, Yixian Liu, Zhichao Hu, Yuhong Liu, Hao Peng