ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2506.15690
14
0
v1v2 (latest)

LLM Web Dynamics: Tracing Model Collapse in a Network of LLMs

26 May 2025
Tianyu Wang
Lingyou Pang
Akira Horiguchi
Carey E. Priebe
ArXiv (abs)PDFHTML
Main:7 Pages
7 Figures
Bibliography:3 Pages
Appendix:2 Pages
Abstract

The increasing use of synthetic data from the public Internet has enhanced data usage efficiency in large language model (LLM) training. However, the potential threat of model collapse remains insufficiently explored. Existing studies primarily examine model collapse in a single model setting or rely solely on statistical surrogates. In this work, we introduce LLM Web Dynamics (LWD), an efficient framework for investigating model collapse at the network level. By simulating the Internet with a retrieval-augmented generation (RAG) database, we analyze the convergence pattern of model outputs. Furthermore, we provide theoretical guarantees for this convergence by drawing an analogy to interacting Gaussian Mixture Models.

View on arXiv
@article{wang2025_2506.15690,
  title={ LLM Web Dynamics: Tracing Model Collapse in a Network of LLMs },
  author={ Tianyu Wang and Lingyou Pang and Akira Horiguchi and Carey E. Priebe },
  journal={arXiv preprint arXiv:2506.15690},
  year={ 2025 }
}
Comments on this paper