#Archived Conversations: reducing friction and ways of encouraging use for User and OAI benefit

1 messages · Page 1 of 1 (latest)

magic sentinel
#

Low hanging fruit:

  • Make it so that "Archive all Chats" button in settings doesn't archive pinned or project chats.
  • Archiving isn't a destructive action so therefore it shouldn't require a confirmation modal.
  • There should be a way to **select multiple chats at once ** (like how you would in a file explorer w/ ⌘ or ⇧ click).
  • use the conversation classifier to go ahead and archive some junk conversations unlikely to be revisited.
  • Better search functionality where by default search excludes archived chats, but offer search operators to allow for deeper searches that may include archived chats.
    • Allow verbatim searches in addition to what I assume is semantic-ish search already enabled.
    • This will make searching much faster if it's skipping archive by default and stronger incentives, UX, or automation existed around conversations archival.

Opinionated Elaborate Idea:

  • Encourage users to not hoard chats or let conversations go too long by automatically (and manually) having chats get marked with a symbolic icon (fa-anchor, fa-bookmark, fa-book, ) to denote chats that are deemed "contextually important"
  • When a "marked" chat is archived, it is distilled/summarized/vectorized then added to some enhanced version of the current memory backed (perhaps a knowledge graph but where each node also contains a reference back to the original conversation or at least a robust summary)
  • The distillation/summarization/embedding process could be done as a batch job with qty and length limits tied to user's subscription tier.
    • It would be vv cool if user's with strong desktop devices to be able to do this on-demand and beyond normal limits w/ on-device embedding/classifier/reranker models.
#
  • Make a desktop/web UI for managing this improved but more interconnected and data dense "User Memory Store".
    • By default obviously it should "just work", but don't let "normal users" decide what good software is. If you build it they will come (the biz/enterprise non-dev power users)
#

I think somebody is gonna have to take a leap of faith and give normies some powerful capabilities before they know what it is or how to use it. We wouldn't be in a situation where enterprise is finding it hard to find productive use cases if the typical office worker wasn't so neutered by the tools they are given and a bar set way too low on what "computer literate" means.

This is a tangent, but I think Archived chats is a good entry point for the knowledge graph and user knowledge vector db

magic sentinel
#

Also it is VERY uncool how arbitrarily restrictive the OpenAI discord is. I can't even add a picture in this thread with something relevant

wooden lily
#

Maybe beacuse they have ai system here and he read it but idk u must read about it somewhere