This post from Andrej Karpathy is interesting:
Something I’m finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it.
He goes on to talk about how he uses Obsidian and a bunch of Markdown files as a knowledge base and gets surprisingly far.
I’ve been using Workflowy as a second brain for years, but have likewise been migrating to local Markdown files + Obsidian + Claude Code. A few things I’ve learned:
- Keep as few repos as possible. Claude does not like to cross repo boundaries.
- For reference type information, have Claude fill in details. Start with an outline, and iterate with Claude.
- If you read a good article that you want to integrate into your thinking, save it. I need to get the Obsidian Web Clipper extension.
- Save images locally. I’ve been using Typora to convert links to local assets, but looks like there is a better way in Obsidian.