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.

Neocky

Linux Colored Shell

Color scheme based of One Dark Pro

How to

1. Open file: nano ~/.bashrc
2. Add to end of file:

PS1="\[$(tput bold)\]\[\033[38;5;9m\]\u\[$(tput sgr0)\]\[\033[38;5;7m\]@\[$(tput bold)\]\[\033[38;5;171m\]\H\[$(tput sgr0)\]\[\033[38;5;7m\]:\[$(tput sgr0)\]\[$(tput bold)\]\[\033[38;5;41m\]\w\[$(tput sgr0)\]\\$ \[$(tput sgr0)\]"

lucanello

# Python Rclone Script with Telegram and InfluxDB Reporting
# - This Script backups your data to any rclone location (see rclone.org)
# - You can execute scripts, commands and database dumps before running a single rclone command
# - It sends automatic error reports and summaries
# - It saves all statistics and operations to an Influx database
# - The data can be used for visualizations in e.g. Grafana
# - You can automate it by using cron

# Created by: Luca Koroll - https://github.com/lucanello
# Last update: 2021-07-04

ucjonathan

(function(options) {
  // Init options if not specified
  options = options || {
    "capturePayload": false
  };

  // Store a reference to the native method
  let oldOpen = XMLHttpRequest.prototype.open;

  var xhrlogcounter = 1;

n-ramdi

// Instagram started setting cross-origin-resource-policy: same-origin when it sees bad referer headers.
// this change leads to ERR_BLOCKED_BY_RESPONSE error and broken images if instagram image is embedded to external website.
// to mitigate this, simple image proxy can be used.
// Steps to install this worker:
// 1. Create CNAME cdn.<yourdomain.com> in your CloudFlare panel
// 2. Create new worker and put the code below into the worker code
// 3. Setup worker route so the worker launches on your cdn. subdomain.
// 4. Modify your image urls from 
// https://scontent-arn2-1.cdninstagram.com/v/t51.2885-15/sh0xxx.jpg 
// to:

soygul

// YouTube API video uploader using JavaScript/Node.js
// You can find the full visual guide at: https://www.youtube.com/watch?v=gncPwSEzq1s
// You can find the brief written guide at: https://quanticdev.com/articles/automating-my-youtube-uploads-using-nodejs
//
// Upload code is adapted from: https://developers.google.com/youtube/v3/quickstart/nodejs

const fs = require('fs');
const readline = require('readline');
const assert = require('assert')
const {google} = require('googleapis');

mdPlusPlus

badblocks -b 4096 -p 0 -s -t 0 -v -w DEVICE [LAST_BLOCK] [FIRST_BLOCK]

	-b block_size (4096 only works up to 16TiB drives, consider using 8192 for up to 32TiB, or 16384 for up to 64TiB. Default is 1024)
	-c Number of blocks which are tested at a time (Default is 64, consider increasing to speed up the process, 64*4096B=256KiB)
	-p num_passes
	-s Show the progress of the scan
	-t test_pattern
	-v Verbose mode
	-w Use write-mode test

ShannonScott

Capture a video of a given X Window with gstreamer.

X Window ID

Get the target window ID with the following:

xwininfo

huozhi

const mouseEventOf = (eventType) => (element, x, y) => {
  const rect = element.getBoundingClientRect()

  const event = new MouseEvent(eventType, {
    view: window,
    bubbles: true,
    cancelable: true,
    clientX: rect.left + x,
    clientY: rect.top + y,
  })

Bartvelp

<?php
/*
Hi!
So this gist is not a very good implementation, it works great in chrome but the seeking is sort of glitched in firefox.
But it does the job and I hope it can help out!

(Please let me know how to improve this)
*/
ini_set('max_execution_time', 0);
$url = "http://mirrors.standaloneinstaller.com/video-sample/jellyfish-25-mbps-hd-hevc.m4v"; //just some sample url, please dont overload their servers