I frequently encounter issues when using ChatGPT-4 to process uploaded PDF files. When I request a summary or translation of a specific page, the response often corresponds to the wrong page—typically around 10 pages off from the one I requested.
To address this, I’ve tried using terms like “the next page” or “the previous page” to guide corrections. However, ChatGPT struggles even with these relative references, often failing to locate the correct adjacent pages based on its own prior responses.
This inconsistency severely impacts the tool’s usability for PDF-related tasks, making it difficult to trust its handling of specific page requests. I hope this issue can be resolved to improve its accuracy and reliability.