When using ChatGPT, the generated output sometimes contains raw internal metadata tags that should not be visible to the user. Instead of clean, human-readable text, the response includes fragments like:
:contentReference[oaicite:36]{index=36}
:contentReference[oaicite:37]{index=37}
:contentReference[oaicite:38]{index=38}
:contentReference[oaicite:39]{index=39}
:contentReference[oaicite:40]{index=40}
:contentReference[oaicite:41]{index=41}
}
:contentReference[oaicite:42]{index=42}
:contentReference[oaicite:43]{index=43}
None
}
}
These look like internal reference markers or debug metadata, and their appearance disrupts the readability and usefulness of the response.
Steps to reproduce:
- Interact with ChatGPT (specific prompt or context may trigger the issue).
- Observe output containing raw metadata tags in the response text.
Expected behavior:
The output should be clean, formatted text without any internal metadata or debug tags visible.
Actual behavior:
Response text contains raw metadata tags such as :contentReference[...]
that appear to be internal references or debug info mistakenly rendered in the final output.
Additional context:
- This issue is related to the text generation or rendering pipeline in ChatGPT, not related to OpenAI API usage or SDK.
- Occurs intermittently during conversations.
- This report was generated by ChatGPT as a test to address detected errors.