Scholarlink

This paper evaluates Large Language Models (LLMs) for static code analysis and defect detection across test sets of increasing complexity. Our findings show LLMs excel with smaller code fragments but deteriorate with increasing complexity. While inconsistent with larger codebases, LLMs provide valuable insights on readability, security, and maintainability that traditional tools miss. We identify key advantages and limitations, concluding that LLMs serve best as complementary solutions to conventional tools, with unique strengths in understanding developer intent and providing contextual recommendations for smaller code units.

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

Leveraging Large Language Models for Software Defect Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ewa Woźny,
  • Jarosław Hryszko,
  • Adam Roman

摘要

This paper evaluates Large Language Models (LLMs) for static code analysis and defect detection across test sets of increasing complexity. Our findings show LLMs excel with smaller code fragments but deteriorate with increasing complexity. While inconsistent with larger codebases, LLMs provide valuable insights on readability, security, and maintainability that traditional tools miss. We identify key advantages and limitations, concluding that LLMs serve best as complementary solutions to conventional tools, with unique strengths in understanding developer intent and providing contextual recommendations for smaller code units.