在代码大模型（Code LLMs）的预训练中，行业内长期存在一种惯性思维，即把所有编程语言的代码都视为同质化的文本数据，主要关注数据总量的堆叠。然而，现代软件开发本质上是多语言混合的，不同语言的语法特性、语料规模和应用场景差异巨大。如果忽略这些差异，笼统地应用通用的 Scaling Laws，往往会导致性能预测偏差和算力浪费。

这项由北京航空航天大学的杨健、国鑫、林静等研究者联合优矿公司和中国人民大学人工智能学院团队完成的突破性研究，发表于2025年12月的arXiv预印本（论文编号：2512.13472v1），是全球首次系统性探索多语言编程训练规律的重要成果。

Community driven content discussing all aspects of software development from DevOps to design patterns. I recently recently compared Java’s REPL scripting environment to Python’s. Many detractors felt ...

Microsoft announced March 2024 updates to its Python and Java extensions for Visual Studio Code, the open source-based, cross-platform code editor that has repeatedly been named the No. 1 tool in ...

Programming language Python is now firmly the second most popular programming language, for the first time knocking Java out of the top two places in RedMonk's language popularity rankings. It's the ...

Microsoft's Visual Studio Code dev teams for Java and Python announced Gradle enhancements for the former and improved Python discovery for the latter. The company updates its associated extensions on ...

C and Java remain the most popular languages in the Tiobe community index, but Python is stalking them and will likely take top spot in the future. According to Tiobe's July 2021 index, the three most ...