This study provides a comprehensive evaluation of how LLM-generated code measures up against human programming across 72 diverse software tasks.

• GPT-4 produced code that was more readable and adhered better to coding standards
• Human programmers created code with fewer security vulnerabilities
• Results show mixed performance across different evaluation metrics, suggesting neither approach is consistently superior
• The research highlights important trade-offs between efficiency, security, and code quality

This work has significant implications for the software engineering industry as organizations consider integrating AI coding assistants into development workflows, emphasizing the need for human oversight particularly for security-critical applications.

Comparing Human and LLM Generated Code: The Jury is Still Out!