Prism Feature Request: Upload Actual Reference PDF Manuscripts for Zero-Hallucination Grounding

ChatGPT Feature requests
1

I am using OpenAI Prism for academic research. While the platform currently integrates beautifully with standard citation keys and Zotero, it relies heavily on metadata. To guarantee absolute factual integrity and completely eliminate hallucinations, we need the ability to import the actual full-text PDF manuscripts of the papers we are referencing directly into the workspace.

Proposed Workflow:

  1. Workspace Reference Vault: A secure section inside the project where researchers can bulk upload the full-text PDFs of their entire reference library.
  2. Google Drive Integration: Allow users to link their Google Drive accounts to directly import, sync, and bulk-upload reference manuscripts. This provides a much smoother uploading workflow for researchers managing massive literature repositories.
  3. Grounded RAG (Retrieval-Augmented Generation): An option to strictly lock the model’s generation to the text contained within those uploaded files, preventing it from hallucinating external details or misattributing claims.
  4. Traceable Citations: Clicking a generated citation key (\cite{key}) should open a side-by-side or split-screen viewer showing the exact page and paragraph in the uploaded source PDF.

This would solve the issue of the model “going off-rails” and make Prism an indispensable tool for high-stakes scientific writing.

2

Thanks for the detailed Prism feedback. We’ll pass along the reference-PDF vault, Drive import/sync, source-grounding controls and precise citation-viewer suggestions. We don’t have a timeline to share, and source grounding would still need careful review for accuracy.