跳到正文
arXiv cs.MA· Dongseok Heo, Junghwan Kim, Bongwon Suh·· 9 小时前AI 评分31

LLM 团队如何检测、纠正并校准其失准成员:Overcooked 环境下的“坏苹果”研究

Bad Apples in the Kitchen: How LLM Teams Detect, Correct, and Calibrate Against Their Misaligned Members

AI 导读

研究向四人 Overcooked 环境注入三种理论驱动的失准智能体,分别施压言语-行动错配、角色入侵和整体进度干扰,并以 3x4x2 设计(n=720 局)拆解检测、纠正、校准三道关卡。几乎所有“捣乱者”最终都会被点名,但只有当归因与情境诊断出现在同一句话中时团队才能恢复;直接做来源归因提示虽改善最难案例,却使无捣乱者基线下降 37.5%。

正文 · 原文

View PDF HTML (experimental)

Abstract:What protects an LLM team when one teammate quietly goes wrong? Most multi-agent LLM work assumes cooperative partners and inspects chat between turns before acting. Real-time embodied cooperation does not allow this pause: chat runs alongside action, so a teammate whose words stay cooperative while behavior diverges can cause damage faster than the team can name what is happening. We study this case by injecting three theory-grounded misaligned agents into a custom four-player Overcooked environment, each stressing a different joint-action layer (speech-action mismatch, role invasion, and aggregate-progress disruption), and decompose robustness into three sequential gates of detection, correction, and calibration through a 3x4x2 design (n=720 episodes). Nearly all trolls are eventually named, but recovery follows only when attribution and situational diagnosis appear in the same utterance; direct source-attribution prompting improves the hardest case only by degrading the troll-free baseline by 37.5%. Robustness, then, depends less on detection alone than on whether team chat combines attribution with diagnosis in a single utterance, suggesting that sensitivity to misalignment should be targeted rather than uniformly raised.
Subjects: Computation and Language (cs.CL); Multiagent Systems (cs.MA)
Cite as: arXiv:2611.22342 [cs.CL]
  (or arXiv:2611.22342v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2611.22342

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Junghwan Kim [view email]
[v1] Tue, 17 Nov 2026 07:51:55 UTC (6,355 KB)

来源:arXiv cs.MA · arxiv.org