Scholarlink

Large language models (LLMs) show promise for structured foresight tasks but struggle with coherence and methodological consistency. This study assesses seven LLMs in generating TRIZ system operator tools (SOTs) across five systems and three prompt types. Performance drops in complex or vague contexts, especially in trend and supersystem areas. Expert reviews and semantic metrics show that breaking down prompts improves output quality. Findings suggest current LLMs need human–AI collaboration and tailored prompting for advanced foresight.

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

Evaluating LLM Performance in TRIZ-Based System Forecasting: A Study Using 9-Windows

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mélusine Caillard,
  • Giacomo Bersano,
  • Gaston Opler,
  • Pierre-Emmanuel Fayemi

摘要

Large language models (LLMs) show promise for structured foresight tasks but struggle with coherence and methodological consistency. This study assesses seven LLMs in generating TRIZ system operator tools (SOTs) across five systems and three prompt types. Performance drops in complex or vague contexts, especially in trend and supersystem areas. Expert reviews and semantic metrics show that breaking down prompts improves output quality. Findings suggest current LLMs need human–AI collaboration and tailored prompting for advanced foresight.