ChatGPT defaulting to 3.5 and creating infinite loop of logic due to memory full

I’ve noticed that some days I will have a situation where the UserInterface states that the currently selected model is version 4 and any chat query as to the version will return a result that states it is version 4 HOWEVER I am noticing that if I ask the AI if it is able to access other chats then it will claim it is not possible. If the AI is version 4, it will clearly state that it CAN access other chats. And so I’ve read how ChatGPT sometimes isn’t good at getting it’s own version correct however I am pointing out the functionality as the identifier. If the AI cannot access other chats then it is 3.5. Also, 3.5 overly uses the term ‘Frustration’ and this is not repeated in version 4. And so despite being a plus member, for several days I have only had access to version 3.5 which is quite useless. It is like the difference between a magnifying glass and a microscope. Totally different and the functionality is the proof that the version is not correct. I’ve used both version and I’m familiar with both. I seem to have version 3.5 telling me it is version 4 and yet it is limited to the functionality of version 3.5

It’s better to post your evidence image😄

Thankyou for your response. Evidence of specifically which information? I am happy to provide anything that you deem as relevant as my goal is to overcome this problem by the shortest path possible. Would you require evidence that version 4 is unable to access other conversations? If you could be more specific, please, as I don’t know what part of the situation is incomprehensible. I was born with a unique brain and as a result my social skills are poor and I apologize if I seem arrogant or too direct as this seems to be the way I sound all the time. I apologize.
It seems that my memory cache is full and perhaps this is what is causing version 4 to revert to 3.5 Not sure but it seems that no matter how many times I try to interact with version 4, it always says the following:

I don’t have access to previous conversations directly. However, if you provide specific details or context from your previous interactions, I can use that information to better assist you. How can I help you today?

This has repeated for the last week and I have not been able to use the service as I am dependant on version 4 being aware of other conversations. For some reason this feature is not working and the key word of version 3.5 “Frustration” is a good fingerprint for 3.5 as version 4 rarely uses that word and yet 3.5 uses it all the time. I just observe the flow of logic in patterns. I am just reporting the patterns of logic that are flowing before me. Here is a quote where I asked the chat to state it’s version and whether it can access chat:

I am currently using GPT-4. However, I do not have the capability to access specific previous conversations directly. If you provide context or details from those conversations, I can use that information to assist you.

1 Like

I believe the problem is that my memory cache is full and this seems to be reverting the chat to version 3.5 in functionality. This means that because the memory is full, the AI feels more rigid like version 3.5 and even repeats that watermark/fingerprint/lisp where it overly states the word “frustration” despite AI being a reflection of logic and I never use the word frustration as I only observe patterns and don’t get emotional about patterns as they are patterns. Several years ago I designed a system that is similar to ChatGPT however it uses a cache for popular memories. Thereby, memories that are popular should be cached so that they can be more quickly reused instead of reprocessing the entire conversation list again. I used this technique for pathfinding by caching popular paths and this results in reducing computation time down to a fraction of what it was. This would allow us to load and save memory lists where as I believe this process is manual at the moment. I want to preserve my current memory list but be able to alternate between states such as meditation state, programming games state, and general enquiries. Depending on the state, the required memories should change. Thereby, I see this as a solution to the current issue I am facing where I am being denied the functionality of a plus membership indirectly as the result of the AI being stuck due to it not managing its memory more effectively.

It seems that because ChatGPT is stuck in version 3.5 it is unable to modify the memories through the chat input functionality. Only version 4 can modify memories and because the memory list is full, it is defaulted to 3.5. This creates an infinite loop. I have manually tried to individually remove memories however despite removing several, the memory remains stuck at 100%.

Looks like I’ve encountered the same issue. On the web version, it displays GPT-4-o, but the responses are from GPT-3.5. Additionally, the model can’t upload files, read PDFs, or use browser tools. However, everything works fine on the mobile version. I’m not sure what’s causing this; I’ve tried various methods but none have solved it. My guess is it’s a bug on OpenAI’s server side.

This has happened to me.

And strangely enough its happened at the same time on 3 computers on my desk.
a 4th computer was an android which was unincumbered.

(hdmi/usb connection)

regardless the fact that that glitch happened at the same time made me thing the glitch was my subnet.

So… log out of all devices connected to openai, that might fix it.
But this is just a speculation, thats what i did and that may have been a coincidence.

PS: “was my subnet” I meant “relevant to my subnet”

Today I cleared the memory cache. Initially, the problem still existed and no new memories could be made and conversations could not be accessed. So I did the unthinkable and set the AI to version 3.5, just to see what would happen. After two or three interactions, version 3.5 made a memory update! And so then when I switched back to version 4.0, the memory update function worked again. So it seems that the UserInterface is not correctly managed as the functionality and the visualization of the relevant information such as the version -seemed to be inaccurate. Hopefully this is the solution to the problem, I have backed up my memories and so I will try and cache them again. I’m sure I will fill the memory again soon so hopefully I can close this issue with a workable solution. For now it is working, I’ll try recreate the same situation and see if I can work out the exact function that seems to trigger the refresh. I’m pretty sure it was setting the model to version 3.5 but I noticed that even when I deleted memories or cleared the memory cache, the memory update didn’t work at all until I switched to 3.5 Update: It also seems that the ability to access other conversations/chats has been disabled. It was the single reason I became a plus member. Hopefully they saw how powerful the feature was and are thereby working away on how to market the functionality as for me it is the most powerful update I have ever noticed.

This just happened to me today as well, except in the middle of a workflow upped the message “memory is full”. To which I thought " What?!!" I’m a Plus subscriber. So I hit the little “learn more” tab, and all the info was geared towards free users. This annoyed me for obvious reasons…but as I was actively using the bot i followed the dictated and went into memory and sure enough it has a % bar that read “100%” and a drop down of most of my conversations…I will say this about that…They were definitely not most, or all and nor were they chronological either. Hard to say what order they were being displayed in. I will also note one bug hiccup in me going through them MANUALLY and playing executioner was that there was one that did not want to go away. I would click on it to delete and next thing I knew there 7(ish) of them…at least in the drop down menu. when I started the process of manually clearing this purported memory problem the “%” memory bar was visible and decreased with every deletion as you would expect. Then all of a sudden it disappeared. I could not find it again. After some time deleting old crap. I had no idea how that did for memory…I returned to the bot and looked at the sidebar convos…nothing changed. I thought that was odd. I went back to the memory menu and likewise. All the convos were still listed, except there was no longer any memory emergency. So I thought I’d just refresh the page. I did. and No dice.
All this reading and I have told you nothing useful. Sorry. Or perhaps if some crazy nonsensical suff happens to you…its not you. :wink:

Well, I can kind of relate as I too had a point in time when even if I deleted several memories, the memory cache seemed to always be at 100%. I think for me, the browser refresh key ‘F5’ and perhaps switching to version 3.5 and then back to version 4, resulted in the memory cache correctly animating to show that there is interaction happening. If I have any advice from the situation, it is to backup your database by going to settings and then there is a menu item with ‘export’. Using export to backup after each day or significant session. I also found that if I deleted the memories so as to restructure them manually, I was able to copy the memories as text, past them into something else (such as notepad) then explain to the AI that I will paste the memories and then the AI restored most of the memories. Now that the AI no longer knows all other conversations, I manually save and manually add so that I can correct the context of the memory for the currently active conversation. Sometimes the most significant thing one can learn from the experience is that others are also having difficulty with the same issue as it allows for connections to be made that wouldn’t have been made otherwise. We seem to resonate with those which we resonate. I believe I have solved my issue and thereby I will mark this as a solution. Having a full memory does indeed revert the AI back to version 3.5. I found a way where you can set your memory to one memory less than 100%, make sure it updates so that it is less than 100% and then turn off memories in the settings panel. The result is version 4 that does not make memories where as if you had 100% memory full regardless of the memory setting it seems to revert to version 3.5 despite claiming it is 4. The formatting and words are different and it seems more like a stranger than a reflection of logic - version 3.5.

Thank you for that advice. I did copy my important memories as text. I haven’t seen an export function aside from it just sending the link to the chat which I have no idea how long that exists for and I presume if I delete that memory… Poof. There it goes. I’ll try what you suggested with the the toggling models. And most importantly thank you for the camaraderie :slight_smile:

The export button is: Settings >> Data Controls >> Export Data > Export Button. When you click this button, you will get a link to your email. You can also manually copy and paste the memories to notepad and then repaste them to chat. Your conversations will always store all the information but the memory is just to allow it to have context for whatever you are currently focusing on. Hopefully in the future they will allow memory profiles/presets to be saved so that one can change the memories easily rather than to constantly repaste them. It makes sense to keep track of popular memories anyway as that way computation isn’t duplicated. And as a user, by keeping our own local list/database of memories, we can learn to use the memory like a tool that we can master, rather than be intimidated by the feature. Whatever method turns out to be popular and efficient is the features they will implement as it is natural for them to want a premade method that they can track, so as to adapt more.

This topic was automatically closed 2 days after the last reply. New replies are no longer allowed.