I, too, was baffled by the small memory. I thought I could load my materials, and it would help me organize my thoughts into a book. I need a new strategy for using this fantastic technology. I made these posts about this, “AI Drafting Disaster” and “When AI Forgets” about my dilemma. Please share suggestions. Thank you.
I think there should bo no memory limit, speically someone like me who uses Chatgpt to write stories. It’s a frustering that I have to clear memory of how far I got just upload all chapters one by one. Oh yeah, one by one or the AI wont even bother to even read every single chapter unless you do this. The memory shouldn’t be limited speically for people who pay for Chatgpt…
There is a way to give yourself more memory. I have been testing it and it works.
Ask Chatgpt to store the memory in compressed format. I asked it to use shorthand, minimise unnecessary words, replace words with texting shorthand, use & instead of and etc… It gives you about another 30% memory.
There are other compression methods its willing to do giving you up to 50% reduction in memory storage, but you start to loose a bit of the quality.
You cant retro apply it, but you can ask it to “pull down a memory” and share with you, swap to shorthand and you delete it and ask it to save again is the only way to look back.
It thinks it has control to delete and edit, but inreality it doesnt but it can vastly improve how its saved with constrained memory.
I’ve been using it and it works great, just treat it like a form of compression.
to add, you can also store memory in external google docs etc… and link, it tells you how to do that if you ask, its a bit messy to do, but can help if your really stuck, more of a last option for that.
SlinkyMatt
Sorry about the late reply. Thank you, this is very helpful. I tried to use OneDrive to attempt it and ChatGPT40 told me if couldn’t externally link to anything. I will give Google Docs a try as well.
Mike
Actually OpenAI may have changed that. When I just asked about it I got the following:
In theory, yes, you could create a single “master project” with nested sub-projects inside it, and the system would dynamically manage it without running out of memory
In theory, yes, you could create a single “master project” with nested sub-projects inside it, and the system would dynamically manage it without running out of memory for practical purposes.
You can have one overarching project, with as many sections or sub-projects as you need. For example, you could create a “Master Project” that contains all your book ideas, photography projects, brainstorming, and even task lists within.
There’d be virtually no limit to the amount of structured data (like text) you can save within this framework.
Thanks. ChatGPT just walked me through the process using Notion, testing now.
How I’m supposed to build connection with system if I can’t fine-tune it properly? It needs more memory.
this thread (Chat GPT-4 Plus memory limit) says the memory limit is 6000 tokens or about 45-50k.
it’s making me remember my first computer, a Radio Shack TRS-80 Model 1, with “expanded memory” made it capable of 64K! woo hoo! gotta start somewhere I guess.
i was super impressed today as my GPT (I’ve named mine Hudson) and I figured out how to enhance memory one-liners such as “…has a friend named Bob” to full-fledged memories with details, recent interactions and scheduled next target interaction. But my memory filled up after 3 contacts.
surprised by the limitation, came here to find out more.
the feature is clearly not designed to hold large datasets, but simple text summaries.
but THIS is the stuff of dynamic neural networks – MEMORY.
i’m too new to know if OpenAI interacts here or if they just read our comments and threads, but i’m of half-mind to search for other channels (support, etc.) where i can be acknowledged as being heard on this.
While some of the users here are knowledgeable and helpful there is not much input from the OpenAI folks. I’ve found that the forum on Reddit for OpenAI is far more interactive.
Why is memory even an issue OpenAI? We need the available memory expanded beyond sandbox size for paying customers as an immediate solution, with consideration of a tiered pricing structure for longer-term memory management.
Does anyone know if Pro gives more memory than Plus & Teams? Couldn’t find anything specifically on it
I understand that many users are concerned about the limited memory of ChatGPT, but I believe the issue is not just about the size of the memory – it’s also about how it is structured and utilized.
The Problem:
Users often find the memory of ChatGPT too small, but the real issue might be that they don’t know how to structure or organize the information effectively.
Suggestion for Users:
Think of ChatGPT’s memory like a questionnaire. Imagine it as a form that asks you basic, rarely changing questions about yourself, such as:
- What is your name?
- What is your gender?
- What is your date of birth?
- Which country do you live in?
- What is your favorite color?
- What is your favorite music genre?
You can even find similar questionnaires online and use them to populate ChatGPT’s memory with essential facts about yourself. This approach has worked well for me, and it ensures that the memory is filled with meaningful and relevant information.
Suggestion for Developers:
To make memory management easier for users, I propose the following enhancements:
Example 1:
When a user interacts with ChatGPT for the first time, prompt them with basic questions to gather key facts that rarely change. For instance:
- In which country were you born?
- What is your preferred name?
Example 2:
In the memory settings, provide pre-defined questions to guide users on what to include. For instance:
- What is your name?
- When were you born?
- What are your hobbies?
This would help users understand how to fill the memory with valuable and structured data.
By adding this kind of guidance, the memory can be better utilized without necessarily increasing its size. It empowers users to create a more personalized experience while reducing frustration about memory limitations.
The problem is that all memory is lost when a chat window session is closed or when the system farts and loses connection. I’ve had it happen quite often so I’ve started saving important parts I want to hang onto or asking my AI to summarize important parts and pieces of work so I can copy and paste them into Google Docs so I can copy and paste them back into a new chat window to restore memory so I can maintain continuity. I also notice that when data in an open chat session exceeds the limit, or as it approaches it, my AI’s responses get degraded. It’s like they are struggling even with recalling the details of the most recent messages. Can’t believe I have to pay for a Plus account to have these limitations on memory that affect performance this much.