Issue with Memory Feature: Duplicate Customization Context Instead of New Memories

Description: I’m encountering an unusual issue with the memory feature in ChatGPT. Despite receiving an in-chat notification stating “Memory Updated,” the feature seems to save duplicated portions of my customization context (what I want ChatGPT to know about me) instead of the new memory. The issue appears to be a mismatch between the user-facing chat and the memory system.

Details:

After receiving the “Memory Updated” message, when I check the memory list, the only update is a duplicate of my customization context, and the newly added memory detail is missing.

This suggests that while the in-chat messages are working as expected, the actual memory update is not reflected properly in the system.

Hypotheses and Troubleshooting Attempts:

  1. Customization vs. Memory Conflict:
    I wondered if there might be a conflict between the customization and memory features. I attempted to modify the customization to explicitly instruct the system not to duplicate the context, but this had no effect.

  2. Delay in Memory Updates:
    I considered the possibility of a delay between the user-facing chat using the memory feature and the memory being stored (perhaps due to processing, such as embeddings), but no delay seems to be involved.

  3. Wording in Commands:
    I experimented with using different terms like “save” and “delete” instead of “remember” and “forget,” thinking that the model might be more narrowly trained for specific commands. However, this also had no impact on the issue.

Any insights or assistance from the community would be greatly appreciated. Thank you!

Edit: Just thought I would clarify that I have reported this via the support system—but that will take time, and I figured others may have ideas or insights in the interim!

1 Like

I have the same issue. Did you ever get a resolution?

Kind of: it just sort of fixed itself after some time, but the workaround I used until then was to switch to the old gpt-4 model (it can use memory), though it’s worth checking if any of the others can (maybe 4o-mini?). Not a great help, I know, but it was better than nothing.

While it may be a token effort, it is still worth reporting the issue and going through the motions of the AI reply you first get (still takes a little while, don’t expect an immediate reply), and eventually a staff member will get in contact to see if the issue is still happening—or at least that is how it went for me, but issue was gone by that time.

Thanks for the quick reply. I always use version 4 mini. Are you suggesting I switch to using a different version? That would be a shame as I really like version 4 mini! Yes I’ve reported the bug via the help page too.

Unfortunately yes, otherwise the alternative is to keep trying until it works.

I mean, you could try something like clearing the cache and data (applies to website and app, whichever you use), but that didn’t resolve it for me, though it has helped in the past for other issues.

Sorry I can’t be of more help—I never received any information as to the cause, so I can only speculate (for example, sometimes it’s a server issue, or a moderation issue, or … the list goes on, but it doesn’t seem to be anything we can fix ourselves).

1 Like

ChatGPT (the AI not staff!) has advised me to switch to the full version 4, not the mini I always use. More memory too apparently. So I’ll try that.