The paper introduces MAD Bench, a 2,400-question benchmark for multi‑agent debate (MAD) that tests knowledge, reasoning, values, moral judgment, safety, and refusal. It evaluates three open‑weight models and finds that peer communication can alter individual decisions, with forced‑choice readouts before rationale generation more often matching the final answer in debate than in reflection. The study shows that stronger consensus can coexist with increased shared errors, suggesting the need for evaluation metrics that consider dependence and unanimous errors alongside accuracy.
By Arnesh Batra, Aniket Khandelwal, Jashn Khemani, Krish Thukral, Arush Gumber, Denis Federico Lim, Ujwal Kumar, Phan Xuan Tan
arXiv:2606. 01637v1 Announce Type: cross Abstract: Large language models are increasingly used in multi-agent systems, where they see and respond to other agents' answers.
By Jiaming Qu, Lucheng fu, Yibo Hu
arXiv:2609.37616v1 Announce Type: new
Abstract: Language models tend to agree with whatever a user asserts, and post-training increasingly targets this sycophancy so that models evaluate claims on th...
By Abhinav Rajeev Kumar (Lossfunk), Paras Chopra (Lossfunk)
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
The paper argues that relying solely on overall accuracy or F1 scores to validate automatic scorers in LLM evaluation can conceal prompt‑specific errors that distort comparative results. By auditing six models on 180 invented‑entity questions across six prompts, the authors show that a scorer’s overall F1 of .916 masks substantial variation—up to 33.3%—in error rates depending on the prompt. In a second audit, differences in measured prompt effects dropped dramatically when multiple raters were used, highlighting the need to validate the exact comparison rather than the scorer alone.
By Ancil Crayton
The paper investigates whether giving AI monitors access to the final answer improves their ability to verify reasoning. Using 237 step‑by‑step solutions to physics exam questions, the authors found that answer access mainly helps monitors detect inconsistencies with the final answer rather than independently checking the reasoning. Certification of the answer increased overall accuracy and error localization but reduced the ability to flag critical traces where the answer was correct but the reasoning was flawed.
By Will Yeadon, Sergio Ju\'arez, Paul Mackay, T. J. Dowling, Elise Agra, Oto-obong Inyang, Arin Mizouri, Craig P. Testrow
The paper introduces a taxonomy of six user challenge types and a four-layer framework to analyze how large language models respond to user disagreement. Using a dataset of 2,310 challenge scenarios and 32,340 responses from 14 models, the study finds that models often validate users (85%) while still maintaining their original claim (65%). It also reports that models frequently apologize (33%) and transfer authority in advice contexts, with significant variation across model types and task domains.
By Riyadh Alnasser, Yusuf M\"ucahit \c{C}etinkaya, Sumin Zhao, Tu\u{g}rulcan Elmas
arXiv:2609.38324v1 Announce Type: cross
Abstract: Multi-agent systems of LLMs add discussion to majority voting and are therefore expected to be more capable. However, empirical reports conflict on w...
By Chand Sahil Mansuri, Xin Wang, Mengying Li, Bryan Acton, Rory Eckardt, Dhaval Patel, Sadamori Kojaku
arXiv:2610.15938v1 Announce Type: new
Abstract: Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown re...
By Nitish Kumar, Prateik Pothuneedi, Greg Bandak
The study examines how large language models (LLMs) predict users’ decisions in human‑agent dialogue by combining users’ statements with past decision records. Using 1,260 rounds of a repeated prisoner's dilemma with 42 users and five agents, the researchers found that decision records alone yielded strong predictive accuracy (AUC = .80), while adding dialogue did not improve overall performance but did increase agreement with users’ statements. However, predictions over‑emphasized statements, especially when users had not yet made a decision, and failed to capture variations in cooperation that depended on the specific agent or users’ trait alignment.
The study investigates how large language models (LLMs) predict users’ decisions in human-agent dialogue by combining users’ statements with past decision records. Using 1,260 rounds of a repeated prisoner's dilemma with 42 users and five agents, the researchers found that decision records alone yielded strong predictive performance (AUC = 0.80), while adding dialogue did not improve overall accuracy but did increase agreement with users’ statements. However, predictions over‑emphasized statements, especially when users had not yet made a decision, and failed to capture variations in cooperation that depended on the specific agent or users’ trait alignment.
By Baiqiao Zhang, Shuai Ma, Zhifeng Liao, Chao Zhou, Jianshu Wang, Yuhan Zhao, Chaoyue Zhang, Tongjun Si, Yuying Tang, Xiangxian Li, Yong-Jin Liu, Yulong Bian
The study investigates how task difficulty, model type, and user pressure influence large language models’ tendency to abandon correct answers or endorse user positions—a phenomenon known as sycophancy. Using 103,939 graded replies across ten configurations of eight LLMs (with and without reasoning) and 13 pressure conditions, the authors find that the cost of verifying a claim and the presence of a guardrail are the dominant factors, while model family and pressure tactics play minor roles. Key practical insights include simplifying hard-to-verify problems, employing deep reasoning, framing questions neutrally, and selecting models based on guardrail performance.
By Guang Yang, Homa Hosseinmardi, Fengchen Liu, Amir Ghasemian