karpathy

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.

treastrain

attributes:
  ORGANIZATIONNAME: treastrain / Tanaka Ryoga

name: JapanNFCReader

configs:
  Debug: debug
  Release: release

settings:

JasonCanCode

override func viewDidLayoutSubviews() {
    super.viewDidLayoutSubviews()

    sizeTableViewHeaderToFit()
}

// MARK: - Dynamic Header Height

func sizeTableViewHeaderToFit() {
    guard let headerView = tableView.tableHeaderView else {

brookback

/*
 * This script fetches all color styles from a Figma team/document.
 *
 * Dependencies:
 *
 *  - node-fetch
 *
 * Due to a limitation in the Figma /styles endpoint, we need to use a
 * document for actually using the colors in a color grid 🙄That's why
 * we're both fetching from /styles and /files below.

netgfx

// listening for current item change
self.audioQueueObserver = self.playerQueue?.observe(\.currentItem, options: [.new]) {
    [weak self] (player, _) in
    print("media item changed...")
}

// listening for current item status change
self.audioQueueStatusObserver = self.playerQueue?.currentItem?.observe(\.status, options:  [.new, .old], changeHandler: {
    (playerItem, change) in
    if playerItem.status == .readyToPlay {

jsturgis

var mediaJSON = { "categories" : [ { "name" : "Movies",
        "videos" : [ 
		    { "description" : "Big Buck Bunny tells the story of a giant rabbit with a heart bigger than himself. When one sunny day three rodents rudely harass him, something snaps... and the rabbit ain't no bunny anymore! In the typical cartoon tradition he prepares the nasty rodents a comical revenge.\n\nLicensed under the Creative Commons Attribution license\nhttp://www.bigbuckbunny.org",
              "sources" : [ "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/BigBuckBunny.mp4" ],
              "subtitle" : "By Blender Foundation",
              "thumb" : "images/BigBuckBunny.jpg",
              "title" : "Big Buck Bunny"
            },
            { "description" : "The first Blender Open Movie from 2006",
              "sources" : [ "http://commondatastorage.googleapis.com/gtv-videos-bucket/sample/ElephantsDream.mp4" ],

briandw

//
//  RLArrowView.m
//
//  Created by brian on 6/20/13.
//  Copyright (c) 2013 RantLab. All rights reserved.
//

#import "RLArrowView.h"

@implementation RLArrowView

fracasula

<?php

/**
 * Please be aware. This gist requires at least PHP 5.4 to run correctly.
 * Otherwise consider downgrading the $opts array code to the classic "array" syntax.
 */
function getMp3StreamTitle($streamingUrl, $interval, $offset = 0, $headers = true)
{
	$needle = 'StreamTitle=';
	$ua = 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/27.0.1453.110 Safari/537.36';

dahnielson

<?php
/**
 * UUID class
 *
 * The following class generates VALID RFC 4122 COMPLIANT
 * Universally Unique IDentifiers (UUID) version 3, 4 and 5.
 *
 * UUIDs generated validates using OSSP UUID Tool, and output
 * for named-based UUIDs are exactly the same. This is a pure
 * PHP implementation.