请问HN:视频展示大型代码库中LLM助手编程的过程
作为一名经验丰富的软件开发人员,我阅读了很多关于人工智能工具如何加快编码速度和软件开发进程的文章。在我个人的经验中,大型语言模型(LLMs)在以下方面提供了帮助:
- 回答问题
- 在空白环境中生成简单代码/框架
与此同时,我在使用LLMs生成一个简单的CRUD应用程序(大约20,000行代码)时并没有取得太大的成功。
我正在寻找的是一段视频,展示一位经验丰富的提示工程师如何使用LLM在一个至少有20,000行代码的代码库中添加一个非平凡的功能,且没有时间间隔或中断。
我所寻找的内容包括:
- 必须用于在一个较大的代码库中添加功能(>= 20行代码)
- 添加的功能不能是孤立的功能(意味着它必须在多个点与系统的其余部分集成)
- 提示的工作量必须少于在编程语言中输入解决方案的工作量,且速度更快
- 任何编程语言/框架都可以
- 任何LLM都可以
- 代码库可以是一个较大的开源项目(因为我假设所有LLM都是在开源项目上训练的,这应该使得LLM的表现更好)
查看原文
As a experienced software developer I read a lot about how AI tools make coding faster and speed up development of software.<p>In my personal experience, LLMs help with:<p>- answering questions<p>- generating simple code/scaffolding in a vacuum<p>At the same time I don't have much success using LLMs to generate code in a simple CRUD application (around 20K LOC).<p>What I am looking for, is a video showing w/o time lapses/breaks, how an experienced prompt engineer uses an LLM to add a non trivial feature to a code base with at least 20K LOC.<p>What I am looking for:<p>- It must be used to add a feature on a bigger code base (>= 20 LOC)<p>- The added feature cannot be a leaf feature (means it must integrate with the rest of the system at multiple points)<p>- The prompting has to be less effort/faster than to type the solution in the programming language<p>- Any programming language/framework is fair game<p>- Any LLM is fair game<p>- The code base can be a bigger open source project (since I assume all LLMs were trained on open source projects, this should make it easier for LLMs to perform)