Hi everyone!
I am trying to generate text for a document based on a pdf guide. The pdf guide explains how pretty much everything should be done and I also have a model to use as inspiration. I am facing some difficulties in getting what I want. Perhaps it is because the pdf is quite big(~100 pages). Anyone got any experience with working like that and wants to share some tips?
Thanks and sorry if this post is in the wrong place, I’m new here.
1 Like
Hi @anton.algeorge and welcome to the community!
Are you using ChatGPT (not API)?
If ChatGPT, it’s bit of a mystery how the document/PDF handling works behind the scenes. If it’s doing any kind of chunking or text compression, or specific retrieval, there is a very high likelihood of instructions not being attended to properly.
In general you want to put your instructions in a prompt explicitly (either in the chat in the textbox in ChatGPT, or as part of a system prompt when calling API). If you are using the new reasoning models like o3-mini, then you want as little text as possible, typically just a few sentences. For gpt-4o you can have larger instructions, but again, they should be more on-point. Including many pages of instructions will likely not yield great results (the model just cannot focus/attend properly).
Yes. I’m using ChatGPT. I sort of got it to work by using the file and also putin in some text when refering to a specific chapter from that document. It is still a bit unpredictible but got the job done. Thanks for the tips.
1 Like