多模态自进化智能体评估时偏好崩溃加剧，跨模态干扰导致策略误选。
Multimodal Evaluator Preference Collapse: Cross-Modal Coupling in Self-Evolving Agents
- 用GPT-4o评估文本与视觉任务，逐步推理策略占48.4%权重
- 真实图像输入下跨模态耦合强度达JSD~0.34，策略选择出现反转
- 自评估模式可免于偏好漂移，97%实验无耦合现象
当AI代理使用语言模型在反馈循环中评估自身输出时，系统性偏差会浮现。我们发现，在多模态场景下，评估者偏好崩溃（EPC）被显著放大。利用GPT-4o评估DeepSeek-chat在文本与视觉任务上的表现，结果显示单一策略（step_by_step）占据48.4%的权重——是纯文本自评估中崩溃程度的3.2倍；而三个视觉领域策略合计仅获9.1%权重。我们揭示了一种新现象：跨模态耦合——某一模态上习得的评估偏好会传递并污染另一模态的策略选择。通过四阶段隔离训练范式，测量耦合系数并记录策略反转：某模态的最优策略在跨模态暴露后发生逆转。在五种评估配置下（共80次独立重复，约35,000次API调用），结合文本代理与真实图像任务的统计验证表明：跨模型评估产生强耦合（JSD~0.19–0.34），真实图像输入提供最一致的方向信号（均值gamma_{T->V}=1.145，gamma_{V->T}=0.937，70% T→V方向一致，Cohen's d=0.56），而自评估则近乎免疫——97%运行结果（N=30）耦合为零（JSD=0.003，d=0.07）。三项方法学消融与多执行器验证确认该效应非结构伪影。我们提出以评估者身份索引的耦合矩阵，发布MM-EPC框架，并识别跨模型评估架构是偏好漂移的主要风险因子。代码与数据：https://github.com/aidless/mm-epc。
原文摘要 · Abstract (English)
When AI agents use language models to evaluate their own outputs in a feedback loop, systematic biases emerge. We show that Evaluator Preference Collapse (EPC) is dramatically amplified in multimodal settings. Using GPT-4o to evaluate DeepSeek-chat across text and visual tasks, we find that a single strategy (step_by_step) absorbs 48.4% of all weight -- 3.2x the collapse observed in text-only self-evaluation -- while three visual-domain strategies receive only 9.1% combined weight. We then demonstrate a novel phenomenon we term cross-modal coupling: evaluator preferences acquired on one modality transfer to and corrupt strategy selection on another. Through a four-phase isolation training paradigm, we measure coupling coefficients and document strategy inversion -- the optimal strategy for a modality reverses after cross-modal exposure. A Phase 3 statistical validation across five evaluator configurations (N=80 total independent repetitions, ~35,000 API calls) with both text-proxy and real-image visual tasks finds: cross-model evaluation produces strong coupling (JSD~0.19-0.34), real-image inputs yield the most directionally consistent signal (mean gamma_{T->V}=1.145, gamma_{V->T}=0.937, 70% T->V, Cohen's d=0.56), and self-evaluation provides near-complete immunity -- 97% of runs (N=30) yield zero coupling (JSD=0.003, d=0.07). Three methodological ablations and multi-executor validation confirm the effect is not a structural artifact. We introduce the coupling matrix indexed by evaluator identity, release the MM-EPC framework, and identify cross-model evaluator architecture as the primary risk factor for preference drift. Code and data: https://github.com/aidless/mm-epc.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。