←back to thread

242 points simonebrunozzi | 5 comments | | HN request time: 0.817s | source
1. sowbug ◴[] No.46236648[source]
Something like this would be perfect for a local LLM assistant.
replies(2): >>46236776 #>>46237124 #
2. sublinear ◴[] No.46236776[source]
Not sure why you're getting downvoted, but I agree at least in principle. There should be some means to index/search this kind of semi-structured text. Summaries are also nice, but not as useful to me at least.

Like the author I also do tagging, but in the real world some notes will eventually slip through the cracks. Even when it's just one, that's probably the one you're looking for. :)

replies(1): >>46237131 #
3. tbeseda ◴[] No.46237124[source]
Agreed. I'm working on a small GUI that just appends to a local .ndjson file. A user just posts with a text box into a feed. Like a one person chat or tweeting into the void. And a local LLM picks apart metadata, storing just enough to index where answers to future questions will be. Then you can use slash commands to get at the analysis like "/tasks last month" or "/summarize work today" etc.
4. anthk ◴[] No.46237131[source]
Either grep or hyperstraier. You don't need an LLM.
replies(1): >>46237694 #
5. nottorp ◴[] No.46237694{3}[source]
A LLM may be able to give you all the paragraphs referring to frobnicating widget X including misspellings and notes not referring to it by name.

It's "AI" right? It could right?