PDF or DOCX - which is easier for GPT to read?

ChatGPT
, ,
1

I use GPT-4 for analyzing and processing my own university course notes. The whole set consists of about 20 docx files.

I found that uploading 20 docx files is not working well. So, I integrated all of the files into one single file, separating the different documents by giving each one a level 1 header, and then produced a meta-index on page one. However, I am not sure if GPT-4 can really read the file well.

So, I converted all 20 docx files into PDF files, then merged all with DEFTPDF DOT com, and again created a meta-index on page one. But then again, when I upload the file and ask about sources, after an analysis, page numbers and even cited text parts are all invented and wrong.

So, what format is better for GPT-4, PDF or Docx? Can it really read one or the other, or does it create for itself a sort of summary after uploading, and thus never can actually read single pages or topics, or understand the structure of a text at all?

2

I am wondering the exact same. Has anyone been able to help with this question?

3

Also wondering this.
In general have folks had better results uploading .docx or .pdf files?
Does length of document or number of documents steer individuals one way or the other.

4

I have found that the context window can easily overflow when uploading a large pdf so I have used stages of refinement to reach more specific answers from a pdf. This is annoying, topic specific, and not scalable. So for now I have found that doing ctrl+f and finding groups of pages referencing something and just handing only that page set is most effective (keep context window small)