brianpetro / obsidian-smart-connections

Chat with your notes & see links to related content with AI embeddings. Use local models or 100+ via APIs like Claude, Gemini, ChatGPT & Llama 3
https://smartconnections.app
GNU General Public License v3.0
2.83k stars 185 forks source link

Chat models can't read notes content #820

Open TroySigX opened 1 month ago

TroySigX commented 1 month ago

When I asked "which of my notes mention subdifferential" and the chat model (llama3.2) answered this image

brianpetro commented 1 month ago

The chat in Obsidian doesn't have a keyword search feature (the Smart Connect GPTs do have a keyword search).

Semantic search (used by the Obsidian Smart Chat) isn't good at that sort of needle-in-the-haystack search problem.

However, the future of Smart Chat will support more actions, like the keyword search action, that are available through the Smart Connect app 🌴

TroySigX commented 1 month ago

Is this use case also a keyword search feature: I create a dummy note with content: "Tom Shane is an American singer." then prompts "Based on my notes, what does Tom Shane do" and it also replies that it can't find any note mention "Tom Shane"

brianpetro commented 1 month ago

Yeah 🌴

Brenzle commented 1 month ago

Hi @brianpetro, I'm loving all the work! Could you give us an ETA for when we can expect the keyword search feature in Smart Chat in Obsidian?

brianpetro commented 1 month ago

@Brenzle The keyword search feature already exists for ChatGPT via the Smart Connect app, and this is the same code that will make it work in the Smart Chat, but it's at least a few months off for the general-release branch.

I've been working on a rewrite of the Smart Chat, which brings the code up to speed with the rest of the architecture/patterns. Part of the reason for doing this is to make room for major updates like integration with the actions provided by Smart Connect.

The Smart Chat re-write is a still a WIP, and I expect it to appear in the early-release version, and also in the Smart Connect app, prior to making it to the general-release branch.

There are a lot of considerations/decisions involved with the above, if you're interested in going deeper then I recommend joining the next Community Lean Coffee meeting where we go deeper into questions like this https://lu.ma/calendar/cal-ZJtdnzAdURyouM7 🌴