After using ChatGPT for some time, I have accumulated hundreds of conversations, some related and others unrelated: questions, research, brainstorming sessions, ideas developed over several chats, and so on. They are typically unorganised.
The problem comes when I remember something I discussed, perhaps months ago, and want to return to that conversation. I often remember the subject or idea quite clearly, but not the chat title, the date or the particular words I used.
ChatGPT’s history search already does one important thing well: it returns a navigable list of matching conversations. The missing part is semantic search.
For example, I recently wanted to find an old conversation where I had developed a model of how humans learn and make decisions and created a diagram.
When I asked ChatGPT directly, it was able to identify several likely relevant conversations But it could not give me a navigable list of those chats - it tries to give me a list of keywords to search on.
So the two required capabilities already exist separately:
-
ChatGPT can understand what kind of conversation I mean and find relevant chat threads
-
History search can return clickable links to the actual conversations.
Please combine these capabilities, I should be able to type into the existing search:
“Find the conversation where I modelled how humans learn and make decisions.”
and have it return relevant chats even if those exact words never appeared in them.
People usually remember what they discussed, not the exact wording they used.