I am hoping not. I am not being one of these people thinking of ChatGPT as a friend, conversing with it about what are being my hopes and dreams, I am using ChatGPT to help me solving problems.
Often I am needing to start new chats because once ChatGPT goes down wrong path too far it cannot finding its way back.
I am with memory off already so this not affecting me, but this is seeming like a very minor “feature” and one which I am believing was already announced long time ago. Is it not just embedding old chats and doing RAG on them?
Yes, I understand what the feature is, but the page was first published several weeks ago. It was unlisted, and I am wondering if there are any changes from the initial version to today’s update.
Thinking a little further, this could be really interesting if that could be enabled individually for selected projects instead of the chat history (chaos), as it would be more like real RAGs with defined purposes.
Of course not mixing all selected projects, more like a RAG for each project.
upper-right: you can now directly delete the chat with a blind kebab menu instead of fidgeting with flaky chat sidebar event javascript that tries to update the DOM on hover events to get to a button:
From the initial reactions it looks like there were quite a few people who were looking to make ChatGPT their personal assistant and buddy. Yeah, great news!
Your chats were always “there”. You already pressed “enter” on the input you previously provided.
There’s no “training on your data” when an embeddings database is used on your past chat turns, as the tech can’t work that way and doesn’t “generate”.
There is only: search results filling the AI context with distraction unrelated to your task every turn. “I’d prefer that with a watercolor style” of another job to confuse your session with global memory enabled.