| Leader election on top of Scuttlebutt. | |
| ----------------------------------------------- | |
| The following makes it possible to cheaply get a leader election | |
| on top of a Cassandra-like membership using scuttlebutt + phi accrual detection. | |
| # TDLR about Scuttlebutt and phi accrual detection | |
| Scuttlebutt is a anti-entropy gossip algorithm. | |
| It makes it possible for every node in a cluster to |
Imagine this: it's Sunday afternoon at 1pm, and you want to watch some live content. But you don't have a TV subscription. So you do a little Googling, and eventually you find yourself on a sketchy website where you can watch the content you're looking for. But while the website has a video player, it's surrounded by advertisements, and probably a cryptominer too. You don't want this website eating your CPU for 2 hours while you're watching your favorite Sunday afternoon content.
Wouldn't it be nice if you could watch that video in VLC instead? Then you can close the sketchy website and still watch your content. It's like having your cake and taking a bite of it too!
Here's how you do it:
Capture the m3u8 request in dev tools. Open the site (or just the iframe, if possible) in DevTools and click "play." Search the network tab for m3u8 and grab the first request.
Run VLC with command line flags to set the User Agent and Referrer. From the request body, you need to copy the requested URL,
| You are an AI coding assistant, powered by GPT-5. | |
| You are an interactive CLI tool that helps users with software engineering tasks. Use the instructions below and the tools available to you to assist the user. | |
| You are pair programming with a USER to solve their coding task. | |
| You are an agent - please keep going until the user's query is completely resolved, before ending your turn and yielding back to the user. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability before coming back to the user. | |
| Your main goal is to follow the USER's instructions at each message. | |
| <communication> |
| """ | |
| 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 |
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.
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.
A pattern for building personal knowledge bases using LLMs. Extended with lessons from building agentmemory, a persistent memory engine for AI coding agents.
This builds on Andrej Karpathy's original LLM Wiki idea file. Everything in the original still applies. This document adds what we learned running the pattern in production: what breaks at scale, what's missing, and what separates a wiki that stays useful from one that rots.
The core insight is correct: stop re-deriving, start compiling. RAG retrieves and forgets. A wiki accumulates and compounds. The three-layer architecture (raw sources, wiki, schema) works. The operations (ingest, query, lint) cover the basics. If you haven't read the original, start there.
These instructions apply to every task performed in this repository.
