Not sure, but it is worse in projects than in normal chat thread. At least in normal chat thread it doesn’t hallucinate. In projects, it is hallucinating.
In normal chat thread it has a clear cutoff of 10000~12000 words. ChatGPT will read until around 12000, but doesn’t hallucinate. They just don’t have information after that point.
In Projects, it is not sure where it starts hallucinating
I have managed to salvage somethings from my project. Basically, once your files goes above 12000 words in projects, chatgpt will hallucinate about that file, they will only read about 2000 words, then fill in the rest with hallucination. Other files in the same projects are able to be read properly(As long as they don’t need to be integrated to 1 file).
If you keep your files under 12000 words, you might be able to do work
I haven’t seen this news on social media. All developers should really retest their chatgpt projects/custom gpts for breakage.
You should all tell your friends in case theirs broke and they don’t know. Multimillion dollars deals or other critical situations might break.