By skyforbes Today, I decided to open a new chat in my project that has all the overviews of my ~23 characters, with equal attention spent on all, and checked whether it could access the memory of my other chats in that project enough for me to be able to discuss the characters in here. It said it could, but that it couldn't easily access memories from outside chats – just how you'd want it to. However, when I then asked it to summarize what it knew about my characters, it proceeded to give detailed information on the character from my backstory chat/project which I've been working on over the past few days, including specific details that I'd only started working out over today and yesterday. Fully separate projects, and yet the memory of the one project seemed to be completely filled with very recent information from that other project, rather than its own.
To me, this feels like it kind of wholly defeats the purpose of projects, since evidently, the memories from these separate projects are somehow still getting jumbled up for me. I told it that it was interesting that it included details from other projects, and it told me this:
"Ah! That makes total sense now — thank you for clarifying. What’s happening is that I’m bringing in some details from related projects and past interactions rather than strictly the memory for this specific current project. […].
For this project specifically, I should really stick to the on-base platoon, barracks life, squad dynamics, and the characters you’ve fleshed out here — so no […] or other external plot elements unless you explicitly want them cross-referenced.
If you like, I can redo the recap cleanly, using only this project’s memory, to give a true picture of the platoon […] as they exist here — no crossover details. That might give you a clearer sense of what’s actually “active” in this project."
Is this something you guys have experienced too – ChatGPT mixing memories between separate projects?