arXiv cs.MA· Weiyi Kong, Shiyang Lai, Jinghua Piao, James Evans·· 2 天前AI 评分58
多 LLM 系统在 200 至 1000 轮交互中出现稳健的语义坍缩
Multi-LLM Systems Exhibit Robust Semantic Collapse
AI 导读
arXiv 论文《Multi-LLM Systems Exhibit Robust Semantic Collapse》指出，多 LLM 交互系统会出现语义坍缩，即在词汇表面存在差异的情况下语义表示系统性收敛，且无需重新训练或更新权重。
正文 · 原文
Abstract:Whether machines can originate novel content has been debated for nearly two centuries, from Lovelace's assertion that no engine can "originate anything" to Turing's question of whether a machine can amplify ideas brought in from outside. Interacting multi-LLM systems, increasingly deployed for autonomous generation, reopen this question empirically. Here we show that such systems exhibit semantic collapse, systematic convergence in semantic representations despite apparent lexical variation, with no retraining or weight updates. The pattern holds across model families and over 200 to 1,000 rounds. Thirteen intervention implementations tested, spanning decoding parameters, prompt design, agent composition, activation engineering, and reinforcement learning, did not reliably restore semantic diversity. Mechanistic analyses suggest that semantic collapse is not solely attributable to the alignment or conformity biases examined here, but is consistent with intrinsic properties of autoregressive generation. Our results point to persistent constraints on the ability of multi-LLM systems to sustain open-ended exploration in closed-loop settings.
| Comments: | 94 pages, 12 figures, 13 tables; includes Supplementary Information |
| Subjects: | Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2605.17193 [cs.MA] |
| (or arXiv:2605.17193v3 [cs.MA] for this version) | |
| https://doi.org/10.48550/arXiv.2605.17193 arXiv-issued DOI via DataCite |
Submission history
From: Weiyi Kong [view email]
[v1]
Sat, 16 May 2026 23:29:42 UTC (3,981 KB)
[v2]
Tue, 1 Sep 2026 21:32:26 UTC (8,478 KB)
[v3]
Wed, 14 Oct 2026 17:23:31 UTC (8,734 KB)
来源：arXiv cs.MA · arxiv.org