That’s what i’m talking about. not every user has the EXACT same problem but all problems lead to the fact that they messed up the memory feature with their last update. the longest memory i have now is nearly 400 characters long. so maybe by probing it i somehow managed to “Fix” it or made a temporary reprieve.
3 Likes
Maybe.But bro is still too low like we had like 1300 words before this words not characters.
3 Likes
true. but like i said, i would recommend probing ChatGPT with every possibility you can think of. that’s what i did. tell it to remember the most ridicolous things and make the memories longer and longer bit by bit to probe where your personal limit is.
1 Like
It’s a bug but they try to keep it quiet… I have the impression that their chat is managed by brainless bots… I had to separate my rp into 50 words, even the relationships of my characters. As well as the layout of a house… a pain but it works…
ha and by the way we can’t share our chats anymore either…
2 Likes
wait what?You can t share your chats anymore?Since when?Don t tell me is from that stupid update on 8th
2 Likes
Let’s get this straight. This is not a bug, but a limitation associated with the cross-reference chat update. It already happened to me when this feature was made available a month ago for me. Just for few days. Perhaps as part of a test group. And a few others as well as I’ve picked up on reddit and here. Some have sent a support ticket. So they knew about it and could have guessed what wave of anger it would cause. Why the limitation? I don’t know, maybe it’s privacy related? It’s a mystery to me, all I know is that it’s not a bug. A lot of people explain it here as well. What’s really infuriating is the pretense and the lack of explanation. If it were up to me, I’d just throw the cross-reference chat feature down the drain (I understand what it’s for, but it can also ruin the context of chat and doesn’t work very well) and keep the original solid working memory. It’s much more important. And… I just don’t like when someone’s bullshitting me if I’m paying for it.
4 Likes
I won’t be able to write my stories or scenes without memory thats the thing 
4 Likes
New response from support. Seems like its not something they are looking into yet
Hello,
Thank you for getting back to OpenAI Support, and we appreciate your detailed observation.
We understand you’re still experiencing an issue where memory only saves very short entries—typically a sentence or two—even after trying all suggested troubleshooting steps. You’ve also pointed out that other users are reporting the same behavior across forums and Reddit, and you’re asking whether this has been officially acknowledged as a bug by the OpenAI team.
At this time, the issue is not listed on our status page, which focuses on large-scale outages or fully degraded functionality. However, your report is valid, and this behavior has been escalated internally for further review. These kinds of nuanced feature degradations may take time to be formally recognized, but user reports like yours are key in helping us track and prioritize them.
To help us investigate more efficiently, please send the following:
- Steps to reproduce the issue
(e.g., the type of content you tried to save, how you triggered memory to save it, and what happened after) - A screenshot/screen recording of what you’re seeing—either the failed memory entry, what is saved, or how it appears in Settings > Personalization > Manage Memory
This information will help our engineering team identify and address the issue more quickly.
Thanks again for your patience—we’ll make sure your report is shared with the appropriate team.
Best,
Justin J
OpenAI Support
4 Likes
yay! so it will get fixed?
1 Like
From the response, I still believe if anything does happen to get it fixed it will still be a while. That’s if they take action
4 Likes
You can literally ask it to save the same exact message multiple times and it will/won’t at random. I’ve got duplicate memories identically worded that had previously failed. I’ve absolutely got “relationships, identity, or personalization” memories saving too. It’s frustratingly inconsistent, very much tied to length, but it isn’t purely a filter thing. I think I mentioned above that for me, the most consistent “fail” is something I’ve saved before (semantically, not identically). Anything that has been memorised before, whether it exists currently or not, is much more difficult. If I say “save this test message”, or “user like the colour yellow” that goes in fine. If I copy a short, neutral memory ChatGPT had created on its own - nada. There may be some kind of cache issue or something as I notice a lot of the time the memory is “Updated Memory” not “Memory Saved”, which has (in the past) been what it shows when updating an existing entry instead of creating a brand new one.
1 Like
I’ve saved all of these in the past two days. I think they all fall into the categories you say are being filtered. System instructions, personality, even some more inappropriately worded entries are in there. But only once. If I delete these, they will not go back. But I can still create new ones if they’re unique-enough (and short).
1 Like
It’s definitely not a filtering issue, when testing the memory over the past couple of days I’ve always used the same paragraph. Most of the time it doesn’t save however I do get the rare moment where it will save the paragraph I’ve entered. It’s very inconsistent though, it only happens like once every 20 attempts.
I’ve also been in contact with open Ai and they’ve also asked me to provide a screen recording of the isssue which I’ve done, so now I’m only waiting for a reply.
2 Likes
Nah, don’t pay their words any mind. Because, what do you expect from a customer service of the biggest AI company in the world? Talking to another human? Even when talking to other human beings on customer support they’d just copy-paste anything and go on without a care about our issues. Let alone a company who makes AI itself? I don’t think you’re talking to a human being, mate. That’s just AI who copy paste your reports and then smooth it over with the power of o3. Don’t believe it if it wasn’t an email. Because I already tried it and got the exact same answers as yours.
3 Likes
I’m almost positive the chat support is literally chatgpt (maybe one of the dumber ones). They literally don’t read what you write, they give the same troubleshooting steps for any issue (“turn on the memory feature”??? Like it’s literally in the screenshots I showed them!). It’s beyond ridiculous that’s the only access we have to support.
5 Likes
Or you guys can report it here. It’s the only place I can find to report anything regarding OpenAI products.
2 Likes
This situation gets weirder and weirder every day. First, the memory issue, then, Chat hallucinating (even more now), completely doing tasks wrong… I can only wonder what is going on.
2 Likes
This is the response from support team that i got today (three hours ago)
We truly appreciate your dedication and clarity you have described your experience
It’s incredibly helpful for us to work to better understand the issue as we work better understand and resolve the issue
We understand how frustrating it must be to see inconsistencies in how memory saved and reflected in memory storage panel, especially after trying so many trouble shooting steps and remaining the patient through out the processes. The level of effort and logical reasoning you have appiled as a both as a writer and data science student has not gone unnoticed and rest assured , we take this seriously
While we are still investigating, your feedback helps in reinforce the importance of the issue and we documented your experience in detail to share with our internal teams
Although we cannot provide daily updates at this time , please know that your report as been noted and we are committed to resolve the issue quickly and thoroughly as possible
If anything else to add or behaviour of chat gpt changes don’t hesitate to reach out
Best regards
Open ai team
2 Likes
I just hope this isn’t a full-on feature and is actually a bug they’re trying to fix, even if it’s not appearing as an error in their status website. I wouldn’t want to waste my days trying to get details of my OCs stuffed inside the AI’s tiny brain that’s now even more strict when catching info.
2 Likes
They need to roll back the syncophancy update on the support chat too
3 Likes