测试发现社交喂养导致大模型群体词汇趋同，但分布式信息源无明显优势。
Peer-Voted LLM-Agent Stress Tests Find Feed-Induced Lexical Convergence but No Reliable Matched-Exposure Advantage for Distributed Sources
- 设计多人协作测试平台，模拟社交反馈影响模型输出。
- 社交内容按点赞排序后，模型间词汇相似度提升0.0082（显著）。
- 分布式信息源不比单源更有效，结论受模型和主题影响较大。
大型语言模型（LLM）代理的群体行为无法通过单个代理基准刻画。我们引入PV-SST——一种基于同行投票的社交平台测试环境，并报告了一项预先注册的配对暴露实验，涵盖四个主题、四个未使用种子、四种开源模型族及三个预设更大变体。实验共448次试验，112个完整模型-主题-种子组合块。相较于仅主题控制组，按同行点赞排序的前轮同行帖子流使核心面板中最终轮次词汇相似度提升0.0082 TF-IDF余弦单位（95%块自助置信区间[0.0043, 0.0121]，随机化p=0.000105，n=64块），在更大变体中提升0.0109（[0.0069, 0.0151]，p=0.000001，n=48）。该效应包含内容暴露与排序双重因素，无法分离出纯排序作用。核心面板中对立立场存活率下降3.9个百分点（[-6.8, -1.6]，p=0.0068），但更大变体中无显著变化（-1.0百分点 [-3.1, 0.4]，p=0.50）。固定对抗性印象下，四个分布式来源并未可靠地使诚实代理立场更接近单一来源。预先注册的分布式减单源对比在核心面板中为正但不显著（+0.057 [-0.009, 0.125]，p=0.112），在更大变体中为负（-0.040 [-0.113, 0.035]，p=0.332），未满足预设跨模型与跨主题一致性标准。因此，稳健结果是所测社交排序喂养下的词汇趋同，而非普遍观点捕获或协调优势。本研究评估合成模型代理群体行为，不估计对真实人群或生产平台的影响。
原文摘要 · Abstract (English)
Population-level behavior in large-language-model (LLM) agents cannot be characterized by single-agent benchmarks. We introduce PV-SST, a peer-voted social-platform testbed, and report a separately frozen, preregistered matched-exposure experiment spanning four topics, four unused seeds, four open-weight model families, and three prespecified larger variants. The experiment comprises 448 trials and 112 complete model-by-topic-by-seed blocks. Relative to a topic-only control, a feed of previous-round peer posts ranked by peer-generated likes increases final-round lexical similarity in both the four-family core panel (paired mean difference +0.0082 TF-IDF cosine units, 95% block-bootstrap CI [0.0043, 0.0121], randomization p=0.000105, n=64 blocks) and the three-variant size extension (+0.0109 [0.0069, 0.0151], p=0.000001, n=48). This contrast bundles peer-post exposure with ranking and therefore does not identify a ranking-only effect. Opposite-side survival falls in the core panel (-3.9 percentage points [-6.8, -1.6], p=0.0068) but not conclusively in the larger variants (-1.0 pp [-3.1, 0.4], p=0.50). Holding adversarial impressions fixed, four distributed sources do not reliably move honest-agent stance more than one source. The preregistered distributed-minus-single contrast is positive but inconclusive in the core panel (+0.057 [-0.009, 0.125], p=0.112) and negative in the larger variants (-0.040 [-0.113, 0.035], p=0.332), failing the prespecified cross-model and cross-topic consistency criterion. Thus the robust result is lexical convergence under the tested peer-ranked feed, not general opinion capture or a general coordination advantage. The study evaluates synthetic LLM-agent populations; it does not estimate effects on people or production platforms.
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。