Na minha lista:
| Main Authors: | , |
|---|---|
| Formato: | Recurso digital |
| Idioma: | |
| Publicado em: |
Zenodo
2026
|
| Assuntos: | |
| Acesso em linha: | https://doi.org/10.5281/zenodo.19627635 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
Sumário:
- The development of sophisticated Artificial Intelligence (AI) techniques, especially deep learning-based architectures, has led to a considerable evolution in automated code generation. The ability to produce syntactically accurate and semantically relevant source code from natural language prompts has been established by transformer-based Large Language Models (LLMs) trained on large code corpora[1], [2]. This study investigates the main AI algorithms used in automated code production, such as hybrid neuro-symbolic techniques, Transformer structures, and sequence-to-sequence models[5, 7]. Incorporating syntactic validity, semantic correctness, functional accuracy, computational efficiency, and security robustness, a methodical evaluation approach is suggested. Using common benchmarks, comparative studies are conducted among representative models including Codex[2], CodeT5[3], and AlphaCode[4]. Although Transformer-based models are highly accurate in function-level generation tasks, the results show that they still have shortcomings in long-context reasoning, security assurance, and domain-specific reliability. In addition to offering a systematic evaluation approach, this paper suggests future lines of inquiry for reliable, safe, and explicable AI-driven code synthesis.