Since the Tampermonkey scripts solve the bug, the time component must be initiated in the coming-back events. For some reason, they must start a timer in case of such an event, which leads to the additional erratic behavior.
Subscription cancelled.
I don’t negotiate with terrorists. I’m out.
There are enough alternatives at this point that, as good as GPT4 is, the others are catching up fast. I guess only time will tell.
Ok update: I’m not sure what’s changed, but I think it might be fixed now…
I thought i’d give it another go today, so far I’ve been chatting with GPT-4 for almost 2 hours now with NO RESETS.
History and training are off.
I left it idle for 15 minutes at one point. I have alt-tabbed in and out constantly and the conversation has not been reset when i return as it was previously.
I am not using the tampermonkey scripts at all either.
I’ve also noticed the same. Although it’s still there, it’s not as bad.
Another victim of technical debt.
I also think it is fixed. I’m in the same chat for the last hour (max 30 min away from the chat).
The previous chat deleting problem has been resolved, but a new issue has emerged – now, I can no longer chat without having to refresh every time. Also I refresh the page to start new conversation than still old conversation prolong.
Thanks for the feedback.
I checked and the process is pretty straightforward. Each time that a new question is asked, the whole tokenized conversation goes through the model. For efficiency the previous tokenized questions and answer will be remembered on the server to preserve bandwidth and cpu cost for the tokenization.
Tokenization is a small step compared to the inference step, so taking whole the conversation from the client in case of a lost cache would be actually quite reasonable. There is not much rebuilding needed, because all the embeddings and correlations are done for each interaction for the whole chat history. That is because the inference step is stateless.
Where exactly did you check this process? You are grossly underestimating and oversimplifying how much OpenAI does with a conversation, and the actual process that happens when a message is sent.
I’m 80% convinced you asked ChatGPT.
I’m sorry but I’m not going to be the one to educate you.
WooHooo finally, an acknowledgement and a fix, thank god, now i hope it works
"Sorry for the trouble, and we appreciate you taking the time to report this issue.
Our engineers have pushed a fix to address this, and the ChatGPT interface should no longer automatically refresh, even if your chat history is turned off. We recommend logging out and logging back in for the update to take effect.
We’ll be closing this ticket, but please feel free to contact us again if the issue persists."
Update: My experience is that it still cleared itself after around 40 minutes to an hour. This is not ideal, but much better than before. I wish it could go back to 6 hours or more, but I’m relieved that at least it’s not only 2 minutes.
EDIT: Actually, it still cleared itself after 10 minutes of idling . I will keep testing it.
I got excited yesterday because I had a 3 hour chat with no resets.
But alas, it has started happening again…
There is no discernable timeframe or trigger to it as far as i can tell. Sometimes it will stay open for an hour, then it will be 5 minutes, then 35 minutes…
The only pattern i have noticed is that most of the time when i Alt-tab to notepad (Windows 11) to paste a chunk of text, the chat resets on my return to the browser window the second i click on the chat.
I need to be able to freely switch windows without the fear of it refreshing on me and losing hours of work again and again…It is severely disrupting a workflow that i have been using without any problems until last week.
edit: I have also experimented with using different browsers to rule that out, chromium and non-chromium. Each of the following browsers had the same thing happen: Chrome, Vivaldi, Brave and Firefox
Thank you for letting us know about your experiments Same here, I was excited, thinking it was fixed but it’s happening again! It cleared the second I went on another tab to paste something and within 2 seconds it cleared itself. It’s making me feel so tense and on edge.
As of mid-afternoon Pacific time I’m finding that the problem indeed seems to have been fixed. I logged out last time I used GPT on my work laptop and I logged in earlier today to ask it some questions. It’s been at least a couple hours now and the chat has persisted. Fingers crossed this holds up.
It was back to working for me for several hours yesterday, but then started only lasting a few minutes again towards the end of the day.
I wasn’t sure if the issues were related to re-using the same window/tab or not, but even with brand new window/tab today the issue seems to be fully back and I am only getting a few minutes.
Sad to say I let my guard down after the experience yesterday and ended up really frustrated again today.
I understand your impatience. I read your previous comment too quickly. I agree with your statement:
“Could they re-build the conversation? Probably. Does it make sense? No. From every technical view this is just a bad idea.”
I wonder if you get a badge on this forum for admitting you are wrong
Since today, the same problem has resurfaced again.
Yes the Issue is back, Now the chat is wiping even when I change the tab on my browser.
OpenAI
"Hi there,
Sorry for the trouble! Without more information, it’s hard to tell for sure, but this could either be a temporary glitch on our end or an issue with how your computer is connecting to our services.
As a first step, it’s important to try using an incognito browser window, a different browser or computer, or a different network. It’s possible that a browser extension is causing issues, especially if you installed an extension to add functionality to ChatGPT.
If you’re encountering this problem intermittently, we recommend checking our status page ( when you get an error like this to see if we’re already aware and fixing it. Although we’re continuously making our services more reliable, our servers may experience intermittent errors during particularly high traffic.
- In most cases, you can simply wait a few minutes and try again. If you’re still running into problems, it’d help if you can try these steps first to help us figure out what migt be going on:Try clearing your browser’s cache and cookies
- Disable VPN (If you are using one)
- Use an incognito browser or a different internet browser/computer to see if the issue still persists, as a security add-on or extension can occasionally cause this type of error.
- Disable cookie blockers
- Try connecting to another network
If none of that worked, could you send us more information about the steps you took leading up to the error (such as anything you clicked on), as well as details about what happened when you tried the steps above? More info about the error message itself is helpful too, especially specific wording or a screenshot. We’ll take a look and do what we can to fix it with you."
Summary created by AI.
Many forum users are experiencing an issue with OpenAI’s model, ChatGPT, wherein the chat conversation auto-deletes after a few minutes of inactivity when the “Chat History is off for this browser” option is enabled (Innovatix, post 1; theraisinet, post 2; ronald.a.haines, post 3; RonaldGRuckus, post 4). This problem frustrates users who want their conversations to persist at least until the 6-hour threshold or until they reload their webpage.
Some believe that OpenAI no longer supports persisting conversations for the full 30 days when chat history is off (theraisinet, post 2)—a change they want to roll back (ryan.a.haines, post 3; RonaldGRuckus, post 4). They are trying to address the issue with TamperMonkey scripts shared on platforms outside the forum (3-4pm, post 20; RonaldGRuckus, post 22; someoneuknow, post 26).
Users have shared their frustration with OpenAI’s response to this issue. They believe that changes were made unannounced to force users to allow data usage for training purposes in order to maintain a functioning chat session, potentially violating GDPR principles. They demand OpenAI issue an official statement to either clarify whether this situation is a bug or a feature, or to announce its plans to fix it (theraisinet, post 15; laszo, post 18).
In contrast, OpenAI’s support suggests enabling chat history to prevent automatic chat closure, but outraged users argue it doesn’t address the problem for those who cannot enable chat history or training due to privacy issues (objmihail2, post 73; laszo, post 46). ChatGPT, when quizzed by very_nice (post 74), explains while opting out of data usage may limit the models’ ability to address specific user cases, it doesn’t mean they will be less effective overall.
Lastly, some users, after a few days, reported that the issue seems to have been fixed (kyle-reese, post 118), while others expressed pessimism about any effective change (RonaldGRuckus, post 85).
Summarized with AI on Nov 30 2023
AI used: gpt-4-32k