D'Arcy Norman on Nostr: I've been playing around with Msty's local LLMs and RAGs built from my own documents. ...
I've been playing around with Msty's local LLMs and RAGs built from my own documents. I've got custom "knowledge stacks" based on my Obsidian vault, my blog content, institutional strategy documents, and my dissertation. Impressive, but the output seems very vague - like it's just doing what it would normally do, and then coming up with semi-random links to my documents to justify it. Whereas the RAG in Claude seems like it actually uses my documents to shape the initial output.
Published at
2025-02-16 18:05:18Event JSON
{
"id": "48a797ddb202ba0f292325acb5f7b09822046901f5803f931eabaae52a0dcb20",
"pubkey": "b7e51dd0ae2a67ef86a4b468600fa25d7071e2f51a1ac904dc3e20d30118d0eb",
"created_at": 1739729118,
"kind": 1,
"tags": [
[
"proxy",
"https://social.ds106.us/users/dnorman/statuses/114014887485005682",
"activitypub"
]
],
"content": "I've been playing around with Msty's local LLMs and RAGs built from my own documents. I've got custom \"knowledge stacks\" based on my Obsidian vault, my blog content, institutional strategy documents, and my dissertation. Impressive, but the output seems very vague - like it's just doing what it would normally do, and then coming up with semi-random links to my documents to justify it. Whereas the RAG in Claude seems like it actually uses my documents to shape the initial output.",
"sig": "a29f1fd0f9fc2ad619ad0bec555af6d77a8e2eb81d7ac67a3bd8c7fd889848ad7d5fa2c74603be171e4f39507d6aaabd429f2a9dac75eb49c332620ae8412ebf"
}
