Open Leizhenpeng opened 7 months ago
真能解决这个问题就能发paper了(no joking) 比如 https://github.com/google/oss-fuzz-gen
我看过一些雏形项目 https://www.bilibili.com/video/BV1Ha4y117rZ/?
直觉上来说,这种模式只适合比较简单的项目。 对于比较复杂或耦合性高的项目,AI得看完所有相关代码才有可能写出正确的代码,尤其是功能性代码。 其次,AI目前不太有自我迭代以及测试代码的能力,就ChatGPT而言,仅支持Python,而且最终写出正确代码的几率不大。
(说实话,这个要是功能真能做好,那估计都得转行产品经理
不知道这个效果如何,回头试试https://github.com/unit-mesh/auto-dev
抬头肯定是简单的改改基础数据,不会很复杂的需求
核心是围绕“测试”去迭代,而不是新增需求
其实,大部分issue都是这种,小步改改。。。。
我现在已经深度使用vscode的Copilot的,把issue问题复制过来,大差不差