There are consistent issues with Codex trying to work on .ipynb (Jupyter Notebook) files, quite common in data science projects.
Most common issue is the file is written in invalid format, also the work seems to be much more laborious for Codex. I’ve moved everything from .ipynb to .py and the reverse once reaching a milestone.
I can imagine this isn’t a simple fix, wanted to provide feedback.
+1 on this. I asked codex to simply revert changes it made to a markdown cell and it ended up deleting a bunch of work I had done in the notebook.
Copilot w/ gpt-5 seems to work fine though. I’m using the VSCode extension
I experience the same issues. Probably because codex has to embed python code into JSON-like format so jupyter notebook can work with it - Like the interactive way of working in JNotebook to inspect datasets . But indeed, might need some more tweaking.
A workaround is to do most of your datascience coding in .py scripts. Simple inspection scripts in JupyterNotebook can be tweaked using the free gitHubCopilot extension in VS code