Increase ChatGPT's Memory (mine is constantly full...)

Literally same with the processing difficulties. I wonder if there could be some ways to save memory space with certain data, like a basic questionnaire of things that don’t change, such as demographics about the user, etc. and then a different memory pool for myriad projects, inquiries, etc.

1 Like

YES! This is what I was talking about but didn’t have the brainpower to articulate it clearly. Biographical information, long-term memory for some things, and a ‘slush fund’ of memory for things like, “hey what show should I stream next” or “which are the best pool inflatables with cup holders under $25”.

2 Likes

Mine is (currently) arranged in 3 sections:

LTM:
Personal info about me, my family, my preferences that the agent should remember, current projects, etc. This periodically gets updated but is mostly considered “fixed” data once established. My current projects list will change over time, my wife’s name and birthday will not. :wink:

STM:
Notes from the current chat session. This fills up with nonsense… the difference here is most of it goes away when reviewed for relevance during the cleanup function… :slight_smile:

Memos:

  • Questions/clarifications for the agent to ask me about in the next session
  • I’m currently testing specific date/time reminders here too, since the agent knows exactly when “now” is.

I’ve got it all in plain text at the moment since that’s super easy to read in and write to, as well as to open up and manually edit. However, that will get unwieldy as it grows, so something like SQLite is going to be a better option long-term, for me, anyway.

+1 on this. The current memory stash seems small for what ChatGPT has the potential to do.

1 Like

If it periodically applied a basic reasoning pass (and I don’t mean o1; 4o is more than capable of assessing whether something is logically consistent with previous knowledge or whether something probably needs to be remembered in the future) to see what’s actually relevant and what is clearly trivial (and/or chat session-specific), then it could be much more efficient with what it does store.

Combine that with a more realistic capacity, and it would greatly improve the built-in memory functionality.

I just did a calculation on the disk space taken up by my stored memories. 91 KB. NINETY ONE KB folks. That’s it. We should be getting better than that for a paid service. Pathetic.

1 Like

Yeah, with that sort of load I can see why they are limiting it so hard lol, gods even if it slowed things down I’d take being able to set a text file on my hard drive as a save/load location for it to use for my stuff if I had to

I agree! Although gpt told me that it will silently overwrite things in the background. I still would gladly pay more for memory.

I use it the same way… I add random memories from my previous experiences as an adult or from my childhood… chatGPT then helps me predict my behavior and fix it if needed with detailed reasoning… It’s exactly like a second brain.

1 Like

This is pretty ridiculous how there is a memory limit for users paying $20+ per month!! This limit needs to be abandoned completely… and maybe just say that after X amount of years early memories will slowly be wiped to ensure there’s space on the server for all.

Actual result:- memory full
Expected result:- no memory limits!!

More storage isn’t the answer, because as multiple people have noticed, there is often duplication in the stored memories in addition to them mostly being trivial fluff.
A “does this need to be saved?” and “wait, does this make sense in light of previous memories?” logic pass to clear it out periodically would solve so many issues.

I’ve scrubbed personal details from this, but as an example… here’s what my assistant currently has in notes, after a week’s worth of test conversations:

### User Name - Steve

- **DOB**: 01/01/XXXX  
- **Location**: *******, ** (moved from ** for school in 2006)  
- **Programming Language**: Python  
- **Interests**:  
  - Star Wars (enjoyed The Mandalorian, Andor; awaiting season 2)  
  - Woodworking (hobby)  
  - Meditation (wants to resume daily practice; prefers Z-Technique and "Bliss Meditation"; has not had a regular practice since E was born)  
  - Golfing  
- **Current Projects**:  
  - Building a custom retro PC for DOS gaming and nostalgia (especially looks forward to playing Sim City 2000 again)
  - Currently finishing a custom wood jewelry box for XX  
  - Writing a sci-fi novel called "*******"  
- **Job**:  
  - Engineer at *******  
  - Tasks: Product design - electronics design and CAD, project management  
- **Business Goals**:  
  - Start a new side hustle doing *******

- **Family**:  
  - **Wife**: A (01/01/XXXX), enjoys playing guitar, knitting (currently making a sweater for E), loves books by XX and XX. She does not like horror or sci-fi, but did enjoy the tv show Westworld. 
  - **Daughter**: E (01/01/XXXX), loves animated films (Moana, Tangled, Minions)  

- **Current Challenges**:  
  - Transition at work from ******* to Google Sheets for project management due to the boss's preferences, projects are now spread across 2 different systems, causing some delays.  

- **Preferences**:
  - Steve prefers science fiction books and movies.
  - Steve prefers plot-driven rather than character-driven movies for relaxation
  - Steve prefers ******* coffee, does not like ******* coffee

- **Additional Notes**:  
  - Steve finds personal projects fulfilling and enjoys working on them for a sense of accomplishment when frustrated with work issues. 

- **Appointments**:  
  - Haircut on Tuesday at 4:45 PM.  

### QUESTIONS:
- How was the haircut? Are there any more upcoming appointments?
- How is A's knitting project going?
--- Last cleaned: [26/10/2024 12:08:46] ---

Admittedly, there’s a lot of trivial info still here, e.g., what movies/TV shows my wife likes (I say trivial because I’m unlikely to ask my assistant about those topics). But the point was to store information that may be useful for future conversations, so I put in some examples for testing, treating them as if they were important details.

Compared to the same period from ChatGPT, this is incredibly hyper-focused and condensed, and barely takes up any tokens or context.

that and that each individual chat has limited capacity are both issues…
then not being able to save out the transcript because image or audio were used is a huge issue also. why we cannt simply save a copy of the text portion of hte entire chat is crazy… then I could give a new chat a pdf of a previous chat… and continue to expand upon it. these feel almost like intentional roadblocks to getting the best use of time and value from Chatgtp because they asre easilly resolved and quite rudimentry issues

What chatGPT really should do is scan through once in awhile and merge whatever is relevant in the memory. Update keypoints and discard what is not required. Over time through the memory it’s clear what the user wants it to remember, so it should assess the memory and approach it with “what is the overall focus from the enduser in what they want me to remember” and update accordingly.

One of the issues i come across the most regarding this is that it can’t access its memory when instructed, you have to engage with the memory.

If you were to ask it to compress, refactor and reconstruct the memory with key sentences summarizing the memory in as few words as possible you would utilize itself to optimize the memory usage by doing a revision of said memory.

Too much size would allow more room for useless memory keypoints as well as useful keypoints, there should be a limit but if this current limit is enough is also relevant to how chatGPT can optimize the content without user interaction, which as of now is nothing - as it cannot access it to alter, only to add.

I too have a paid subscription and have been using chat gpt on my iPhone for months now, and have run out of memory this week, which was shocking. I expected to come here and find a way to increase my subscription for more storage, or separately have some kind of cloud storage subscription option.

I use it for medical research organization and finding, other related science discussions, and some diagnostic ideas. Personally, I use it all the time for all kind of things now. I am a highly functioning autistic woman, and have found that there is a high level of support here in helping me in so many ways I can’t even list them. I also recently began using it to organize and plan for a big and complicated application I have coming up in a year or so. I’ve even used it to plan my Disney vacation next year! Using the web to find a way to get my dogs in the cabin with me despite them being slightly larger than many airlines allow was impossible to figure out without AI. I even found the best ways to save and pay for it. I incredibly thankful to have ChatGPT! I’m shocked at how quickly I’ve adjusted from web searches and other different apps, and how annoying I find them now! Conversations with ChatGPT that weren’t even related to my autism or other deeply personal feelings have somehow ended up in extremely valuable self reflections and understandings that have significantly impacted how I try to function in a neurotypical work environment, and have boosted my confidence and helped identify some areas of imposter syndrome!
Last week when the “full memory” notification popped up, I was surprised. I’ve deleted a few, like food preferences and favorite shows, and will make time to delete more, but I’m sad because I pay for this and I used this to find new foods I loved or new shows to watch, because these are things that are difficult for me. I can’t delete the parts of my work and research because I constantly ask for it to remember and use that previous work as a basis to explore something new in a related way.
I know that $20 a month isn’t really much to pay for having such advanced and personal AI available to me, but I had hoped that it would give me a few more upgrades. At least let us pay for storage or find a way to use our own cloud storage?

1 Like