创建

登录 ReadmeX

或

了解我们如何处理你的信息: 隐私政策

ReadmeX

陪你读懂 AI 动态

今天,想先了解什么?

选一份动态总结,或直接聊聊你的问题。

社区和关注动态总结只属于你自己。

资讯

Reddit 用户称 Codex 反复全量测试拖慢任务

AI 总结

一名 Reddit 用户称,Codex 会反复运行整个应用的测试套件,包括未修改的代码,因此基础编程任务耗时逐渐增加。用户表示,要求 Codex 只运行针对性测试并解释额外测试的必要性后,平均任务时间从两小时以上降至约 10–15 分钟;这一说法尚未得到独立验证。

深度见解

该案例显示全量测试可能占据编程智能体的大部分任务时间,并提出以针对性测试作为缓解方法,但所称效果尚未独立核实。

So codex was taking longer and longer to do basic tasks. I paid attention to it and realised it was running a full test suite, multiple times, of the entire application i'm developing, even when I ask it to make a small change to one part of the program. It spends forever checking its own work and running tests on things it hasn't even modified. If it finds one issue while testing, it fixes the issue and then runs another full test suite.

It reports how many individual things it's tested. It would say it's done thousands of tests, found one issue, then ran the exact same thousands of tests, found another issue, another thousand tests...

Tell it to do targeted tests only on the specific thing that is worked on, and justify why any other additional tests are needed. Now each task I set doesn't take 2 hours anymore going round in circles. 10-15 minutes on the average task and it's done.

Try it.

#Codex CLI#OpenAI

原文约 1 分钟读完阅读原文
群聊

还没有评论,来说说你的看法。