Advancing Code Generation: Insights into Large Language Models

Abstract

Large Language Models (LLMs) have demonstrated significant potential in automated code generation, aiding in tasks ranging from writing software prototypes to industrial control systems programming. These models offer unprecedented capabilities in generating and refining code. Yet, they face challenges like limited explainability, lack of execution guarantees, and the need for specialized support in niche domains such as Programmable Logic Controllers (PLCs). This paper comprehensively explores LLMbased tools across various fields, including scientific research and industrial applications. We propose novel frameworks like LLM4PLC, which integrates user feedback, external verification tools, and fine-tuning techniques to ensure the correctness and safety of generated code. Additionally, we investigate the role of LLMs in enhancing code understanding, demonstrating that embedding these models in developer environments can significantly improve productivity and task completion rates. Our findings highlight the promise and limitations of LLMs in code generation, with results showing substantial improvements when leveraging structured verification pipelines and context-aware tools.