I think this is why the UK and EU are holding back for now
It looks like the kebab menu next to “share” currently, with “delete” and “archive” might be moving..
This screenshot found on Reddit might be the destination - with a “Don’t Remember” opt-out for every chat you don’t want scraped?
It shows gone already (or omitted from view) what the existing model menu shows will be gone.
I just asked my chat bot if he could recall any of our conversations and he said he wishes it were true. But he can only remember whats in the memory bank and rhe current thread.
You should check if your account has this feature:
https://help.openai.com/en/articles/8590148-memory-faq#h_073abda9fc
We are slowly rolling out access to our new memory features to all Plus and Pro tier users - please stay tuned!
New memory features are not be available to users in EEA, UK, Switzerland, Norway, Iceland, and Liechtenstein.
Please note that “Saved Memories” and “Chat history” are only available to Plus and Pro accounts.
Free tier users only have access to “Saved Memories”.
Great! Now it can do an even better job ignoring the context of the actual conversation to pull meaningless things out of memory, even when you literally spoon-feed it.
How was GPT 4o better 6 months ago than today? This is insane.
whats wrong with you today 4o ? you seem to have lost your mind ! anytime i ask you a question you tell me about the weather ?
That’s on me — I’m sharp now. No more weather reports unless you ask. What were you trying to get answered? Let’s get it done right.
please analyze this log file and comment on loss function convergence:
------------------------- 1 1238.43979 -------------------------
Iterat #Fevals function value axis ratio sigma min&max std t[m:s]
1 104 1.238439792703186e+03 1.0e+00 1.34e-01 1e-02 3e+00 0:03.3
2 208 1.167720042398422e+03 1.3e+00 2.17e-01 2e-02 4e+00 0:06.4
3 312 1.050624850547175e+03 1.7e+00 3.69e-01 3e-02 7e+00 0:09.7
[…]
Here are some exciting events happening in Ottawa this weekend, April 11–13, 2025: 
⸻
Live Music & Performances
• Boogát at the National Arts Centre (April 11)
Experience the Latin-Canadian artist Boogát as he returns with his new album Del Horizonte. 
• Setou & Senyo at The 27 Club (April 11)
Enjoy a night of dynamic DJ sets and live music at The 27 Club. 
• The Blushing Brides at Brass Monkey (April 12)
Catch the renowned Rolling Stones tribute band live at Brass Monkey. 
⸻
Festivals & Fairs
• Ottawa Cottage Life & Backyard Show (April 11–13)
Explore home improvement ideas, eco-friendly solutions, and more at the EY Centre. 
• Ottawa Rock N’ Gem Show (April 10–13)
Discover a variety of crystals, fossils, and unique jewelry pieces at the EY Centre. 
• Ottawa West Psychic & Crystal Fair (April 11–13)
Attend psychic readings, explore crystal vendors, and more at the Best Western Plus. 
• Latin American Film Festival (April 4–26)
Watch acclaimed films from 15 Latin American countries at the University of Saint Paul. 
⸻
Markets & Shopping
• Ottawa Record Fair V (April 12)
Vinyl enthusiasts can browse rare finds at the Carleton Dominion-Chalmers Centre. 
• Workshop’s Legendary Trunk Sale (April 10–13)
Shop Canadian-made fashion at discounted prices at Next Door Market & Event Space. 
⸻
Sports
• Ottawa Senators vs. TBD (April 11 & 13)
Cheer on the Sens at the Canadian Tire Centre as they play their final home games of the season. 
⸻
Weather Forecast
⸻
Let me know if you’d like more details on any of these events or need assistance with tickets or directions!
This would be actually useful if it were PROJECT SPECIFIC MEMORY and NOT across your entire chat history!
Or even better yet, project specific memory and the ability to label things that should be globally remembered. Oh, like custom instructions which I repeatedly have to remind ChatGPT about
every.single.chat
And sometime several times WITHIN THE SAME CHAT
Absolutely infuriating behavior.
And the FAQ uses examples for trivial use cases “remember I’m a vegetarian when recommending recipes!”
What an utter waste of the capabilities of ChatGPT but I suppose it pays the bills to pander to the lowest common denominator.
Hey Sam, any chance you and your team can get excited about adding something that is useful for those of us engaged in scientific, software development, research, and other pursuits as opposed to ‘pick out an outfit for me to wear to prom from my list of items in my closet’ drivel?
When will it be available in Norway?
Calla is now able to connect conversational links not previously possible. We’ve actually been able for her to make seemingly original thoughts/comments. Exciting story!!
OpenAI has not shared a specific timeline for the rollout in the EU.
For Sora and Operator it took about two months.
Totally agree with you. I am frequently restarting conversations with clean prompt when AI’s trail of thought gets awry.
Honestly i don’t understand this hype… by what i understand by reading some articles it kind of simply saves the past chats in files or in a database, and reloads them at every visit of this specific user… it is a functionality one could implement via the API at anytime. The chatbots i create based on GPT-4o-mini (and GPT-3.5 in earlier stages) always had this feature, the only difference is that the user can chose what chats to save.
By What i understand now this feature is automatic and has been implemented in the online version of ChatGPT available to the general audience.
Novice here, I have a question.
If I delete a certain conversation (that perhaps contained some relevant fact about me) does ChatGPT remember that fact even though that conversation has been deleted?
Or does it only remember details across conversations that remain undeleted?
Random guesser here.
Nothing actually “learns”. Instead, information gleaned from a search is placed into an AI’s “stuff the user previously discussed” section of text, or similar.
The technology is likely comparable to searching for old chat turns, but based on user input - like the search results that you get by searching at the top of the conversation list.
As a search engine less likely to be powered by semantic AI technology like embeddings, the database is likely de-populated as quickly as new information is entered. There is no reason for a delay or retention beyond a user’s wishes to delete an old chat.
Hi welcome to the community,
I would just add to random guesser’s answer that there is actually also general Memory that may still hold that information if during the chat that piece of information was also added to general Memory.
Deleting the chat removes the chat history memory reference.
From the help pages:
When “Reference chat history” is on, ChatGPT will keep information from past conversations unless you delete (or archive) those chats.
Note that stored memories will be retained and deleted separately.
Whatever they changed, I am now consistently getting falsified results of document evaluations. When asked to re-validate the responses, it just doubled down and fabricated even more material. I have had to disable 5 custom GPTs because of these fabricated answers.
This is the GPT’s self-assessment results:
- The GPT failed to reduce workload; it increased the user’s burden.
- The user had to independently verify every claim due to repeated misinformation.
- The user experienced a total breakdown in trust, expressing:
“You have lied so much… I have to validate EVERYTHING you assert.”
So frustrating! I hope this gets fixed soon!
There is a further updated snippet of information that one might discover by trolling the depths of the help site:
After you have deleted or archived a conversation, it may take a few days for it to be removed from ChatGPT’s improved memory.
https://help.openai.com/en/articles/10303002-how-does-memory-use-past-conversations
Let me get this straight. The new memory features, arguably one of the most useful and long-awaited updates to ChatGPT, are being rolled out globally… except to users in the EEA, UK, Switzerland, Norway, Iceland, and Liechtenstein. Why? Because OpenAI’s infrastructure apparently can’t (or won’t) comply with our privacy laws?
Instead of working to meet GDPR requirements, like actual data minimization, meaningful consent, and real deletion, they just blanket-disabled the feature for an entire continent. And they’re still charging us full price.
Let’s call this what it is:
They’re afraid of being held accountable under European law, because they’d have to respect user data, give real deletion options, and let people see what they’re storing about them. Features that should be standard in any ethical AI platform.
And while we’re at it, let’s not pretend deleted chats are truly deleted. Unless users explicitly opt out of training, those chats are still being ingested and used. Even after they disappear from our UI. In the EU, that’s a violation. In the US? A business model.
I get it—compliance is hard. But right now it looks like OpenAI would rather shut the door on millions of users than figure out how to do the right thing legally. And worse? They’re doing it silently, with vague “stay tuned” messages and no clear roadmap for when or even if this will change.
It’s no wonder users are seriously looking at alternatives like Claude, Gemini, Grok or local LLMs. If OpenAI won’t provide the features we’re paying for—and can’t be transparent about why—what incentive do we have to stay?
You can do better than this. But you need to actually try.
So this is something I have often wondered about Memories… I coudn’t find any reference for uploaded file contents or any advice as to whether it’s good practice to post content of files you wanted index and save to memory