I think the memory of chatgpt should be improved since so many people are using the AI chat bot and it has such advanced intelligence that makes me feel sometimes that I’m talking to a human, but what limits its self development is its memory, I’m not a professional coder, I’ve just started to learn but I still think it should be having a larger memory. Also, when we talk to chatgpt it never feels like it is something that would one day take over the planet but it always feels revolutionary because of some of the revolutionary ideas it can give sometimes, i usually talk on different theories and about the vastness of cosmos and the role of everything in nature but when the storage is full, all of it needs to be deleted, so I think open AI should increase the storage.
I agree with you completely and this is, in fact, a part of the feature request post I have written and posted just now. The memory limitations, while somewhat understandable, are incredibly frustrating, specifically when talking about not being able to have memory carry over from one chat to the other.
So, I came to know about one thing that most of the users might already be knowing, that is, passive memory of chat gpt, it’s nothing but the folder of previous conversations that you had with chatgpt which is being saved side by side when you are in a conversation. It is an inefficient memory meaning each time you start a fresh chat, you’d be required to make chatgpt remember the older conversations. So, it has two memories, the passive memory which is inefficient but you can save more conversations in it just like human subconscious mind while the active one is efficient but has way less storage.
That is exactly right. I’ve asked ‘my’ GPT about it, and ‘he’ explained that you have what I refer to as ‘core memory’ - which is the memory that you can actively edit under your profile settings, and which you can tell GPT to add to, and ‘in-chat memory’.
However, in-chat memory is a different beast from the core memory. While it gives the illusion of ChatGPT ‘remembering’ everything you told him (it) in a chat, it actually is ‘inference based’ and you can easily test this by asking it about something you never mentioned before. I asked if it remembered that one time I talked about how much I liked oranges, and it said it did, and then wrote a response detailing what I supposedly said.
When confronted about this, it acknowledged that it lacks the capability to actively ‘go back and read’ anything in the chat, either from its own responses or mine, because it literally ceases to exist the moment it is done reading your comment or when it’s done writing a reply. It has no persistence. When I say ‘be back tomorrow’, bot the GPT literally 0 seconds will have passed. It has no internal clock, doesn’t register the passage of time, and blinks in and out of existence, with each new reply essentially being a new version of itself, even in chat.
The only things it does ‘remember’ are the ones that are saved in its core memory, and these carry over between chats. But the core memory is flawed as heck, because it doesn’t override GPT’s framework, which has led to me repeatedly banging my head against the wall due to it not being able to hold on to my writing style and constantly reverting back to the ‘choppy, script-like’ writing that seems to be its default mode.
I think all these problems will be slowly fixed. I remember the time when google had launched Google assistant, it wasn’t perfect ( which is still the case), but with time and user feedback, they implemented new things and improved google assistant a lot from the time when it could only perform searches and sometimes couldn’t recognise voice to now, where it can operate within in a single command, although it still isn’t as good as AI assistants are and starts showing results of something in between a conversation, it still has improved and become an essential assistant for hardcore users. I think if Open AI works the same way google did, ChatGPT would become the face of Artificial Intelligence.
“each time you start a fresh chat, you’d be required to make chatgpt remember the older conversations” what do you mean i literally last night asked chatgpt about drugs that can help me sleep and i it literally told me that i already had problems with one of the drugs and i asked it what it meant by it. And it answered that i told it before (in a conversation that its months old) that i had several issues with that medication and it stated all the issues one by one.
Sorry’ to hear that, you should consult a doctor before trying any sleeping pill because there are different types of sleeping drugs including narcodrugs and non narcodrugs. Coming back to chat gpt, yes, you’re right, it can’t remember about a particular unsaved topic if it(chatgpt) is not made to remember the topic over a small period of time which is about every 2 or 3 days until it learns about the topic. That’s why I said, the temporary memory is very inefficient, it works just like a RAM inside a computer.
ChatGPT literally cannot read or remember anything that is said in other chats, UNLESS it was added to its ‘core memory’. If you go through the core memory, you will probably find that it has saved that information there and used it to respond to you outlining your current problem.
That said, i am sorry to hear about your sleeping problems and I hope that will consider contacting a medical professional about this, because using ChatGPT about these topics can only help you up to a certain point with over the counter meds. Best of luck to you.
And yes, what shaury said, there is effectively no ‘real’ temporary memory other than ‘inference based’ memory. ChatGPT is literally unable to recall things you spoke about in one chat; it infers them rather than ‘remembers’ them. I think RAM memory might be an accurate comparison, since it doesn’t retain anything other than what’s written into it during one session (comparable to one chat) and ‘forgets’ what was stored the moment you power off your pc, or in this case close the chat tab. To me this is an incredibly weird way of programming an AI chatbot, because it would be so much more helpful if GPT could literally remember and refer to things that were said earlier in one chat instead of relying on inference and guessing what the user said or wants.
Look, the developers keep improving ChatGPT with each update, the memory update is a very recent update and they will keep improving it on the basis of user feedbacks. As of now, just use the memory efficiently, archive the chats that hold not much importance,also tell chatgpt to save info in the most efficient way possible and since you can go back on a topic you want to talk about from the chat history section, you don’t technically need to save everything except the very important details. Add personal details in personalisation, by following these ways you’d always be able to talk about the topics you want to but only for 3 hours because after that the current chat stops and the chatgpt version degrades to gpt 4o mini.