代码大模型 / AI 编程 - arXivDaily 专题

2601.12262 2026-01-21 cs.SE cs.CL 81%

Environment-Aware Code Generation: How far are We?

环境感知代码生成：我们有多远？

Tongtong Wu, Rongyi Chen, Wenjie Du, Suyu Ma, Guilin Qi, Zhenchang Xing, Shahram Khadivi, Ramesh Periyathambi, Gholamreza Haffari

机构 * Monash University（墨尔本大学）； Southeast University（东南大学）； eBay Inc.（eBay公司）

专题命中代码生成：code generation(title,abstract)；分类 cs.SE、cs.CL

AI总结本文提出环境感知代码生成的研究，通过引入VersiBCB基准测试，探讨了数据、参数和缓存三个适应轴，揭示了LLMs在环境特定代码生成中的挑战与改进机会。

Comments ICSE 2026

2601.13118 2026-01-21 cs.SE 79%

Guidelines to Prompt Large Language Models for Code Generation: An Empirical Characterization

为大型语言模型生成代码的指南：实证分析

Alessandro Midolo, Alessandro Giagnorio, Fiorella Zampetti, Rosalia Tufano, Gabriele Bavota, Massimiliano Di Penta

专题命中代码生成：code generation(title,abstract)；分类 cs.SE

AI总结本文通过实证分析提出10条提示优化指南，帮助开发者改进代码生成提示，提升LLM辅助软件开发工具的效果。

2601.13015 2026-01-21 cs.SE 79%

MeltRTL: Multi-Expert LLMs with Inference-time Intervention for RTL Code Generation

MeltRTL: 多专家LLMs结合推理时干预用于RTL代码生成

Nowfel Mashnoor, Mohammad Akyash, Hadi Kamali, Kimia Azar

专题命中代码生成：code generation(title,abstract)；分类 cs.SE

AI总结 MeltRTL通过多专家注意力和推理时干预提升LLM生成RTL代码的准确性和效率，实现96%的可综合性和60%的功能正确性，无需微调模型。

2507.03160 2026-01-21 cs.SE 79%

Assessing Small Language Models for Code Generation: An Empirical Study with Benchmarks

评估小型语言模型用于代码生成：基于基准的实证研究

Md Mahade Hasan, Muhammad Waseem, Kai-Kristian Kemell, Jussi Rasku, Juha Ala-Rantala, Pekka Abrahamsson

专题命中代码生成：code generation(title,abstract)；分类 cs.SE

AI总结本研究评估了多种小型语言模型在代码生成中的性能，发现紧凑模型在效率与性能间取得平衡，但进一步提升需转向更大模型。

Comments 17 pages, 10 Tables, 57 figures. Includes benchmarks and multilingual evaluation. Submitted to the Journal of Systems and Software

2601.02076 2026-01-21 cs.CL cs.AI 62%

Deferred Commitment Decoding for Diffusion Language Models

延迟承诺解码用于扩散语言模型

Yingte Shu, Yuchuan Tian, Chao Xu, Yunhe Wang, Hanting Chen

机构 * Peking University（北京大学）； Huawei Technologies（华为技术）

专题命中代码生成：code generation(abstract)；分类 cs.CL、cs.AI

AI总结本文提出延迟承诺解码方法，通过基于不确定性的策略提升扩散语言模型解码质量和效率，实验显示在多个模型和基准测试中效果显著。

2502.04829 2026-01-21 cs.LG cs.AI 62%

Optimistic Gradient Learning with Hessian Corrections for High-Dimensional Black-Box Optimization

乐观梯度学习与Hessian修正用于高维黑箱优化

Yedidya Kfir, Elad Sarafian, Sarit Kraus, Yoram Louzoun

专题命中代码生成：code generation(abstract)；分类 cs.AI、cs.LG

AI总结本文提出OHGL算法，通过结合乐观梯度学习和高阶梯度学习，提升高维非线性优化性能。

Comments We develop a black-box optimization algorithm that learns gradients with neural models and can be applied to solve non-convex high dimensional real-world problems

Journal ref AAAI2026

2601.08857 2026-01-21 cs.SE cs.AI 62%

Revisiting Software Engineering Education in the Era of Large Language Models: A Curriculum Adaptation and Academic Integrity Framework

重新审视大语言模型时代的软件工程教育：课程适应与学术诚信框架

Mustafa Degerli

专题命中代码生成：code generation(abstract)；分类 cs.SE、cs.AI

AI总结本文提出一个理论框架，分析生成式AI如何改变软件工程能力，并推动教育模式向过程透明度转变，以应对大语言模型时代教育与实践的不匹配问题。

Comments The conclusion part is not complete. The correct and complete paper will not be uploaded

2601.04700 2026-01-21 cs.CL 57%

PRISM: A Unified Framework for Post-Training LLMs Without Verifiable Rewards

PRISM: 一种无需可验证奖励的统一后训练LLM框架

Mukesh Ghimire, Aosong Feng, Liwen You, Youzhi Luo, Fang Liu, Xuan Zhu

机构 * Arizona State University（亚利桑那州立大学）； Amazon Web Services（亚马逊网络服务）

专题命中代码生成：code generation(abstract)；分类 cs.CL

AI总结 PRISM通过结合过程奖励模型与自我确定性，实现无需真实标签的稳定训练和提升LLM测试性能。

Comments Added open-sourced github url

2601.12995 2026-01-21 cs.CL 57%

Graph Reasoning Paradigm: Structured and Symbolic Reasoning with Topology-Aware Reinforcement Learning for Large Language Models

图推理范式：基于拓扑感知强化学习的结构化和符号推理用于大语言模型

Runxuan Liu, Xianhao Ou, Xinyan Ma, Jiyuan Wang, Jiafeng Liang, Jiaqi Li, Tao He, Zheng Chu, Rongchuan Mu, Zekun Wang, Baoxin Wang, Dayong Wu, Ming Liu, Shijin Wang, Guoping Hu, Bing Qin

机构 * Harbin Institute of Technology（哈尔滨工业大学）； State Key Laboratory of Cognitive Intelligence（认知智能国家重点实验室）； Tianjin Normal University（天津师范大学）； Pengcheng Laboratory（鹏城实验室）

专题命中代码生成：code generation(abstract)；分类 cs.CL

AI总结图推理范式通过拓扑感知强化学习实现结构化和符号推理，提升大语言模型的数学推理和代码生成能力。

2507.01436 2026-01-21 cs.HC cs.AI 57%

Challenges & Opportunities with LLM-Assisted Visualization Retargeting

基于大语言模型的可视化重定向挑战与机遇

Luke S. Snyder, Chenglong Wang, Steven M. Drucker

机构 * University of Washington（华盛顿大学）； Microsoft Research（微软研究院）

专题命中代码生成：program synthesis(abstract)；分类 cs.AI

AI总结本文探讨了大语言模型在可视化重定向中的应用挑战与机遇，通过对比两种方法评估LLM在代码生成和适应中的性能及局限性。

Comments 5 pages, 3 figures, 1 table

2601.13815 2026-01-21 cs.AR 50%

From RTL to Prompt Coding: Empowering the Next Generation of Chip Designers through LLMs

从RTL到提示编码：通过LLMs赋能下一代芯片设计师

Lukas Krupp, Matthew Venn, Norbert Wehn

专题命中代码生成：code generation(abstract)

AI总结本文提出基于LLM的芯片设计教育平台，帮助初学者通过RTL代码生成和提示编码实现可tapeout的芯片设计，展示了LLM在芯片设计教育中的应用效果。

Comments Accepted for presentation at the 2026 IEEE International Symposium on Circuits and Systems (ISCAS 2026). Proceedings to be included in IEEE Xplore

1. 代码生成 11 篇

Environment-Aware Code Generation: How far are We?

Guidelines to Prompt Large Language Models for Code Generation: An Empirical Characterization

MeltRTL: Multi-Expert LLMs with Inference-time Intervention for RTL Code Generation

Assessing Small Language Models for Code Generation: An Empirical Study with Benchmarks

Deferred Commitment Decoding for Diffusion Language Models

Optimistic Gradient Learning with Hessian Corrections for High-Dimensional Black-Box Optimization

Revisiting Software Engineering Education in the Era of Large Language Models: A Curriculum Adaptation and Academic Integrity Framework

PRISM: A Unified Framework for Post-Training LLMs Without Verifiable Rewards

Graph Reasoning Paradigm: Structured and Symbolic Reasoning with Topology-Aware Reinforcement Learning for Large Language Models

Challenges & Opportunities with LLM-Assisted Visualization Retargeting

From RTL to Prompt Coding: Empowering the Next Generation of Chip Designers through LLMs