Instead of all chats being a sum of user input and chat replies, there should be a clear distinction between the two and the ability to remove wasted responses.
For example, without even asking for assistance, while just trying to add details to be used for reference, we get massive responses to questions we never asked. This “fluff”, becoming part of the chat, as well as the responses where we ask chat to exclude the replies. Leading to “memory full” issues.
If the chat was isolated as input, with a selection to ask questions and get answers that didn’t become part of the logged chat, that would save a lot of memory.
Since AI would likely give the same response, orbs better one later, why even save chat responses at all? Just regenerate a new response, if it is even needed.
In that same respect, if, within a conversation, we ask to have a prior edit made, and it was confirmed. Then the details about asking for the edit, the approval and the repeated correction, are no longer needed. Just dispose of them. This goes with the idea that there should be some separate “interaction layer”, to be isolated from the users, “refined input layer”.
For ease of future use, chat-gpt should offer to export a compressed version, which only chat-gpt could understand, as a form of archive. As opposed to saving all the bloated html formatted text and token data, as a whole. Preferably to a linked online account, where chat-gpt can easily access that data, for future inclusion.
I will offer a specific example of, “unsolicited fluff”, that keeps wasting my “memory”, which I can’t get rid of, or control.
I tell chat-gpt that I am going to give it some story details, which I want to be checked for continuity and conflicts.
Then it basically repeats what I just said and adds offer for other aid. (Repeating could have been reduced to “okay”. Not responding to the solicitation to further aid, should have just removed that solicitation reply.)
I tell it some details, in length, and it replies with banter about how creative my dynamics are. Though I never asked for opinions. It also starts spewing out potential dialogue, also not asked for. Then it starts to horribly summarize the submitted data, quite prematurely, also not being instructed to do so.
All that useless fluff, consuming nearly half my memory. Instead of just listening, until an actual question is asked. Or only responding with the requested, prior aid, of noting conflict or continuity issues.