Managing PDF Documents in ChatGPT 4o Sessions: How to Avoid Memory and Token Issues?

I uploaded 3 PDF documents in ChatGPT 4o, and began asking it about specific aspects discussed in these documents. I always use the same chat window, meaning it’s always the same session. I’ve noticed that in the following days, as I continue writing in the same window (session), at some point it seems like it no longer references the information contained in the uploaded documents. In fact, I copied a portion of text from one of the PDFs and asked if it could find it within the uploaded PDFs, but it couldn’t. I had to re-upload the same document to then receive confirmation that the text was found in that document.

The questions I wanted to ask are:

  1. Does this happen because I have exceeded the token limit for that session, causing it to start erasing older information to make room for new data?

  2. If I use GPT 4o with a longer context version, would I be able to have longer sessions and therefore “more memory”?

  3. If I use GPT 4o mini to upload documents within a session, theoretically allowing me to upload longer documents, and then later switch within the same session from GPT 4o mini to regular GPT 4o, could I leverage its greater cognitive ability while also saving tokens for that session, or does it not work that way?

  4. How can I solve this issue and avoid having to re-upload documents repeatedly to get consistent answers?

1 Like