Use of Large Language Models in Code Generation

Authors: Sanjay Sunil, Sudha D

DOI: https://doi.org/10.37082/IJIRMPS.v14.i3.233096

Short DOI: https://doi.org/hbz96h

Country: India

Full-text Research PDF File:   View   |   Download

Abstract: Out of nowhere, big language machines started reshaping how software writes itself, moving far beyond old rigid rules into smart, learning-based engines. These new models - like GPT-4 [1], Claude 3.5 Sonnet [2], DeepSeek-V3 [3], and Gemini Pro [4] - can turn plain speech into working code, jumping between coding styles and tech areas without skipping a beat. Now arriving - LLM-powered coding tools are no longer lab curiosities but live parts of daily developer work. GitHub Copilot, fueled by OpenAI's Codex, sees regular use among coders; research points to faster output under specific conditions. Yet weaving large models into dev pipelines stirs concerns: how solid is the output, can it be trusted, what happens to the coder's craft when machines suggest half the lines. Even with big steps forward, problems remain when using large language models to create code. Though GPT-4 gets functions right most times, its output runs efficiently less than half the time - just 45.4%, while functionally correct in 83.1%. Efficiency lags far behind accuracy, data shows. Code written by machines often clashes in style; research finds mismatches appear in two-thirds to nine-tenths of cases versus human work. Performance shifts wildly based on field - one robot task might succeed 95 out of 100 tries, another drops as low as one in five, depending on which model is picked. This work dives into how well large language models create code, looking at what they can do and where they struggle - covering various systems, coding tongues, settings. Findings from up-to-date research come together here, showing where things stand right now while pointing out gaps that need attention down the road.

Keywords: Large Language Models, Code Generation, Natural Language Processing, LLMs, AI-Assisted Programming

Paper Id: 233096

Published On: 2026-05-02

Published In: Volume 14, Issue 3, May-June 2026

Share this