I've encountered a strange issue with GPT where it completely ignores any content from my PDF file after page 27. The document is 30 pages long, and while it acknowledges the total page count, whenever I try to get a summary or details from the last three pages, it just circles back to discussing earlier content. It's puzzling because I've heard that GPT can handle much larger documents, yet it seems to be stuck on this one reference point. Has anyone else experienced this?
3 Answers
If your PDF doesn't have images, consider converting it to Markdown or plain text. The added formatting in PDFs can confuse models like GPT. If you really need to work with the full document, try pasting the text directly into the input box instead for better results.
You might want to try splitting the PDF into smaller chunks. Sometimes the model has a limit on how much text it can process at once, and files longer than 10 pages can be tricky. It's worth a shot!
Exactly! The way LLMs work with their context makes it hard for them to keep track of everything in longer PDFs. In my experience, breaking it up helps a lot.
The issue could also be that it's managing data in tokens rather than straightforward page counts. This means it might be misjudging how much content it can handle after a certain point.
Yeah! Even if it's long, having GPT summarize a tighter version could be useful before asking more detailed questions.