Scholarlink

<p>This study analyzes a decade of human writing (2014–2024) across PubMed, Wikipedia, and Stack Exchange to examine if it has converged towards machine-predictable patterns. Using perplexity and Binoculars metrics from three language models (OPT-125&#xa0;M, GPT-2, and Llama-3.2), combined with non-probabilistic metrics (average sentence length and type-token ratio), we find that human writing has become less linguistically diverse and more machine-predictable over time. This trend predates the widespread adoption of generative AI, suggesting it is not solely due to direct LLM use. While several factors like standardization efforts may contribute, the recent acceleration of convergence suggests potential reinforcement from AI-human interaction, highlighting a bidirectional influence between human behavior and AI systems in shaping written communication.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Human writing and machine patterns: analyzing a decade of convergence

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Eunsuk Chang

摘要

This study analyzes a decade of human writing (2014–2024) across PubMed, Wikipedia, and Stack Exchange to examine if it has converged towards machine-predictable patterns. Using perplexity and Binoculars metrics from three language models (OPT-125 M, GPT-2, and Llama-3.2), combined with non-probabilistic metrics (average sentence length and type-token ratio), we find that human writing has become less linguistically diverse and more machine-predictable over time. This trend predates the widespread adoption of generative AI, suggesting it is not solely due to direct LLM use. While several factors like standardization efforts may contribute, the recent acceleration of convergence suggests potential reinforcement from AI-human interaction, highlighting a bidirectional influence between human behavior and AI systems in shaping written communication.