I fine-tuned a local LLM on my Obsidian second brain to explore the future of augmented intelligence. Here's a walkthrough of fine-tuning DeepSeek R1 Llama 8B, explore embedding visualizations, and my reflections on what this means for the future of personal AI.
I also go into how I use AI today and what we need for true Augmented Intelligence.
Key topics:
Fine-tuning LLMs on personal knowledge bases
Visualizing note embeddings and knowledge clusters
Creating targeted QA datasets
Using Llama Factory for efficient fine-tuning
How I use AI today
The future of augmented intelligence
🕐 TIMESTAMPS:
0:37 - My Obsidian Second Brain
1:00 - Creating Embeddings
2:03 - Exploring Embeddings & Note Clusters
5:43 - Building QA Dataset
6:17 - Creating QA Pairs
7:04 - Fine-tuning Process
9:44 - Was it Actually Useful?
10:10 - How I Use AI in my Second Brain Today
10:45 - How I Use AI for Augmented Intelligence
12:07 - Future of Augmented Intelligence
Code lives here: https://github.com/bitsofchris/deep-learning/tree/main/code/06_obsidian-rag-fine-tuning