I've recently started using ChatGPT Plus, and I'm getting the hang of it, but I've noticed some limitations. I saw someone mentioning that they uploaded thousands of self-help books for therapeutic purposes, but when I upload even one PDF, it seems to misunderstand things and doesn't retain info well. Another user mentioned that ChatGPT struggles to remember PDFs over 4,000 words. I've been trying to save my chats and memories, hitting the cap often, but it feels like I can't just return the chats as a file and pick up where I left off. I'm curious about how the person with thousands of PDFs managed it. Was ChatGPT once able to handle more text? Have any settings changed? Sometimes it feels like each new chat session is a blank slate, while other times it retains some information. I'd love an easy-to-understand explanation!
4 Answers
I’ve uploaded about 20 documents under 20,000 words each, and it works pretty well for me. Just keep the lengths in check, and it can manage it fine!
Honestly, ChatGPT isn't the best for this kind of deep interaction. It tends to draw from general knowledge, leading to confusion at times. Check out NotebookLM instead. It has a 2,000,000 token context window, which is perfect for large PDFs. Plus, it's free right now, so totally worth trying out!
It’s true that while you can upload a lot of text to ChatGPT, the limits depend on the account type. Plus accounts should handle more than 4,000 words comfortably. If you’re storing a lot of chats, consider making 'summary memories'. Break up the text and capture essential insights to have an easier-to-manage reference later.
Good idea! I was worried about hitting the overall cap, so I’ll keep that in mind while managing my summary files.
Does ChatGPT even use RAG for attachments? I thought it just includes them in the context directly.
So, from what I know, plus users have a context window of about 32,000 tokens, which is roughly 128,000 to 144,000 characters. However, this includes space for system prompts and instructions, so it's best to keep some room free. I tackle big files by breaking them into chunks of around 80,000 characters. That way, I generally avoid issues and keep things manageable!
That makes sense! I started using summary files too. They help a lot, especially if I need to highlight key points without overloading the system. I'm optimistic they'll improve memory handling in the future!
I heard version 4.1 has a massive 1 million-token context, but that's good to know for how to handle big files!
Thanks for the recommendation! I’ll definitely look into it!