analiza aiqbench
This arXiv paper studies how LLMs change answers during multi-agent debate compared with independent reflection, finding that debate can raise consensus and agreement while also increasing unanimous wrong answers and inter-agent error correlation. For AI readers, the practical takeaway is that agreement among model agents is not evidence of independent verification, so multi-agent setups need evaluations that measure error dependence alongside accuracy. The findings are based on open-weight models and a specific benchmark, so generalization to other systems should be checked before drawing firm conclusions.
Who Convinced Whom? Tracing How LLMs Change Their Minds in Multi-Agent Systems
De la arXiv
Text integralarXiv:2610.35930v1 Announce Type: cross Abstract: Multi-agent debate (MAD) aims to improve language-model decisions through the exchange of arguments, yet agreement after debate need not reflect independent corroboration. We investigate how peer communication influences individual decisions and collective outcomes through behavioral and internal interventions. We introduce MAD Bench, a 2,400-question benchmark spanning knowledge and reasoning, values and moral judgment, and safety and refusal, and evaluate three open-weight models using a round-matched independent-reflection control.
Among answer-changing responses, a forced-choice readout before rationale generation matches the eventual answer more often in debate than in reflection (65.4% versus 22.3% on Qwen3.5-9B). Message ablations reveal model-dependent effects of peer answers and reasoning, while attributing identical text to the receiver rather than another agent substantially changes its influence. For the tested source descriptions, stated track record affects answer preferences more strongly than professional-status cues across all three models.
Attention masking produces different changes in decision margins across the three checkpoints. On fully parsed, reference-labelled non-numeric questions, debate yields small estimated accuracy improvements over initial majority voting, while increasing the rate of unanimous wrong answers by 3.1-7.4 percentage points relative to reflection. Longer debates show persistent consensus and greater correlation between agents' correctness than reflection. These findings show that stronger consensus can coexist with more shared errors, motivating evaluations that assess dependence and unanimous errors alongside accuracy.
Title:Who Convinced Whom? Tracing How LLMs Change Their Minds in Multi-Agent Systems
Abstract:Multi-agent debate (MAD) aims to improve language-model decisions through the exchange of arguments, yet agreement after debate need not reflect independent corroboration. We investigate how peer communication influences individual decisions and collective outcomes through behavioral and internal interventions. We introduce MAD Bench, a 2,400-question benchmark spanning knowledge and reasoning, values and moral judgment, and safety and refusal, and evaluate three open-weight models using a round-matched independent-reflection control. Among answer-changing responses, a forced-choice readout before rationale generation matches the eventual answer more often in debate than in reflection (65.4% versus 22.3% on Qwen3.5-9B).
Message ablations reveal model-dependent effects of peer answers and reasoning, while attributing identical text to the receiver rather than another agent substantially changes its influence. For the tested source descriptions, stated track record affects answer preferences more strongly than professional-status cues across all three models. Attention masking produces different changes in decision margins across the three checkpoints. On fully parsed, reference-labelled non-numeric questions, debate yields small estimated accuracy improvements over initial majority voting, while increasing the rate of unanimous wrong answers by 3.1-7.4 percentage points relative to reflection.
Longer debates show persistent consensus and greater correlation between agents' correctness than reflection. These findings show that stronger consensus can coexist with more shared errors, motivating evaluations that assess dependence and unanimous errors alongside accuracy.
Multiagent Systems (cs.MA); Artificial Intelligence (cs.AI)
arXiv:2610.35930 [cs.MA]
(or arXiv:2610.35930v1 [cs.MA] for this version)
https://doi.org/10.48550/arXiv.2610.35930 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 324 cuvinte
Subiecte conexe

Current AIs out-persuade professionals in lab settings but (probably) not in the world
LessWrong · 5 oct. 2026
Thinking Less to Simulate Better: Intuitive Prompting Improves LLM Agents Simulating Individual Social Media Reactions, Including Unfamiliar Content
arXiv · 28 sept. 2026
From Language Models to World-Acting Systems: Progress and Limits of Agentic AI across Digital, Social, Virtual, and Physical Environments
arXiv · 7 sept. 2026

'Sophisticated' AI swarm attacks are months away, OpenAI warns: What experts say businesses must do
ZDNet · 31 aug. 2026