During OpenAI’s GPT-5 launch event, they demoed the model’s ability to fix real bugs in production code. Live on stage. In their own repository. The kind of demo that makes CTOs reach for their credit cards and engineers nervously update their resumes. There’s just one small problem: the fix they promised to merge “right after the show” is still sitting there, unmerged, three and a half months later.
The issue, presumably the PR (linked at the top of the issue because of reference).
Look at the code change. It gets inputs and loops over them and seems to do an in-place fixup. But the code indent is wrong, and it even changed the function definition of the unrelated next function. In Python, the indent-logic-significance language.
I assume they briefly showed the code on stage. Even then it should have been obvious to any developer. py file, messy indent, changes unrelated function.
Please correct me if this is the wrong PR.
Based on what they showed in the demo, that’s not the PR. gpt made more changes than what’s in that PR and also modified a different transform function. My guess is they never actually pushed the actual commit from the demo or made a PR.
Does not appear to be the correct MR. Comments on the issue allude to “they never pushed it” so sounds like there never was an MR. Watching the announcement where they demo’d it, it wrote much more than is in that MR. Not to defend OpenAI, I hate vibe coded solutions that add so many useless comments.
Write. Readable. Code.
What’s with the comments? “Hi mom”
So “LuminaX-alt” is an AI?