Thomas B. TomTom101

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

Routing tips for VPNs on OS X

When VPNs Just Work™, they're a fantastic way of allowing access to a private network from remote locations. When they don't work it can be an experience in frustration. I've had situations where I can connect to a VPN from my Mac, but various networking situations cause routing conflicts. Here are a couple of cases and how I've been able to get around them.

Specific cases

Case 1: conflicting additional routes.

In this example the VPN we are connecting to has a subnet that does not conflict with our local IP, but has additional routes that conflict in some way with our local network's routing. In my example the remote subnet is 10.0.x.0/24, my local subnet is 10.0.y.0/24, and the conflicting route is 10.0.0.0/8. Without the later route, I can't access all hosts on the VPN without manually adding the route after connecting to the VPN:

	"""
	The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp