I’m a professor and long-time researcher integrating Indic wisdom and Western science for peer-reviewed publications. I’ve been using ChatGPT extensively to write and revise a ~10,000 word manuscript for the Journal of Management, Spirituality & Religion (JMSR), and my experience with the recent GPT-4o model has been extremely unstable.
This manuscript involves:
-
Sanskrit glossary consistency
-
APA 7 reference checks with DOI validation
-
Nested ontology models (guṇas, bhāva, Śakti)
-
Dozens of citations, cases, leadership theory comparisons
-
Explicit instructions not to change key terms, sections, or ontology structure
Since switching to GPT-4o, I’ve had to:
-
Upload the manuscript 11 times
-
Watch it fail to fetch content from uploaded files correctly 15+ times
-
Repeat explicit instructions multiple times (e.g., “don’t change my pyramid model”)
-
Manually correct glossary and reference inconsistencies despite prior approvals
This is not a complaint about GPT-4o. It’s a request for OpenAI to support power users who are working on:
-
Books
-
Academic papers
-
Curriculum development
-
Long-term research synthesis
Please allow an opt-in “Stable GPT-4-turbo Mode” or a Long-Form Writer/Researcher Modethat prioritizes memory, stability, and instruction adherence over speed or casual conversation.
These tools are amazing—but they’re becoming harder to trust when doing serious work.
If others are experiencing similar issues with deep research or long-form work, please upvote or comment. Let’s build momentum for this feature request.
—Prasad Kaipa
PS: > I’ve spent hundreds of hours working on this research and manuscript. It’s been hard to continue due to memory lapses, overwriting of user-approved content, and version unpredictability. A dedicated long-form writing or “researcher mode” would help hundreds of professionals like me.