I am going all in on projects and been under the impression that projects have shared memory. Further, I was under the impression that project shared memory can be tagged to segment memory and limit which GPT instances within a project can access what topics. Today, I learned that as with most things there is a limit on shared project memory. When I tried to view the Shared Project Memory I realized that shared memory and personal memory are all one. If this is correct I would have security and go-forward design concerns. Can someone point me to any documentation or content that can help me understand projects at a deeper level? For example, project file limitations, possibly using files as shared project memory as opposed to personal memory - looking forward to feedback and collaboration.
Welcome to the forum!
The official information published by OpenAI is this FAQ
Also see:
If someone knows of more official OpenAI info I would be interested.
That is one of the best ways to learn about new features. For some new features such as ChatGPT Teams, Canvas and ChatGPT Windows Desktop app, this is how I learn most useful information beyond the FAQ.
Just continue to ask questions in this or a new topic and share what you find and others might join you in your endeavor. If you plan to make this a focal point of such consider changing the title and align the content for such.
Thank you for replying - I did check out that article you linked to and really is not very deep. I also asked ChatGPT to help me debug itself but it was also surprised to learn that project and personal memory is one and the same - my hunt continues…wish I could design projects.
You could always use the OpenAI API to create something, or even a personal GPT.
I have toyed with the idea of a personal GPT - I think all GPT should be personal to start you decide if you want them to be part of a greater community. Long ago, not in a galaxy so far away, I created the first distributed intelligent agent platform - it was called KYMA Atlantis and one of the attributes was canonical interaction - meaning that the agents started from the most restrictive modality and then as the user, or their orchestrator you could pull back their restrictions - but I am highly intrigued by private or personal GPT for my own use or as a tool for government or advanced research projects.