Did anyone else notice that GPT-6/codex uses (inline) Python much more aggresively?
一位用户在测试GPT-6/codex时发现，该模型在处理复杂问题时更频繁地使用Python，包括内联Python和Python脚本，甚至通过SSH注入Python。这种行为增加了“潜在问题输出”的范围，使得人工审查变得更加困难。该用户已添加了一个本地SKILL来缓解此问题，并想知道其他用户是否也遇到了类似的情况。
- 发布
- 2026年9月7日 09:39
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
时间以 UTC 显示
更多信息
I am testing GPT6 and noticed a few things. A bunch of problems or 'more difficult situations' often cause it to generate python scripts or inline-python or even inject it via SSH and other ways.
Sometimes to avoid editing a file in the project by a few edit calls, it just generates one script to do it. And the 'Approve for me' function in Codex kind of always accepts it.
For me as a human who is still reviewing what is happening, it had become increasingly difficult since the 'potentially problematic output' surface increases a lot through this. I added a local SKILL to tone this done, but I wonder if others have similar issues.