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.

ninehills

otmb

import SwiftUI
import WebKit

enum WebViewError: Error {
    case contentConversion(String)
    case emptyFileName
    case inivalidFilePath
    
    var message: String {
        switch self {

alexandremcosta

defmodule Downloader do
  @moduledoc"""
  Download streams of bytes from URLs.
  Useful to transfer large files with low RAM usage.
  
  ## Example with `ExAWS.S3.upload/3`
  
  ```elixir
  url
  |> Downloader.stream_body!()

y0ngb1n

Docker Hub 镜像加速器

国内从 Docker Hub 拉取镜像有时会遇到困难，此时可以配置镜像加速器。

Dockerized 实践 https://github.com/y0ngb1n/dockerized

1️⃣ Docker daemon 配置代理（推荐）

参考 Docker daemon 配置代理

undoZen

module.exports = {
  h5: {
    miniCssExtractPluginOption: {
      filename: 'css/[name].[hash].css',
      chunkFilename: 'css/[id].[chunkhash].css'
    },
    enableDll: false,
    webpackChain(chain, webpack) {
      const hashRule = (type) => {
        chain.module.rules.get(type).uses.get("urlLoader").get("options").name = 

yycking

import JavaScriptCore

extension JSContext {
    subscript(key: String) -> Any {
        get {
            return self.objectForKeyedSubscript(key) as Any
        }
        set{
            self.setObject(newValue, forKeyedSubscript: key as NSCopying & NSObjectProtocol)
        }

scrapehero

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Written as part of https://www.scrapehero.com/how-to-scrape-amazon-product-reviews-using-python/
from lxml import html
from json import dump,loads
from requests import get
import json
from re import sub
from dateutil import parser as dateparser
from time import sleep

jswny

Prelude

I. Preface and Motivation

This guide was written because I don't particularly enjoy deploying Phoenix (or Elixir for that matter) applications. It's not easy. Primarily, I don't have a lot of money to spend on a nice, fancy VPS so compiling my Phoenix apps on my VPS often isn't an option. For that, we have Distillery releases. However, that requires me to either have a separate server for staging to use as a build server, or to keep a particular version of Erlang installed on my VPS, neither of which sound like great options to me and they all have the possibilities of version mismatches with ERTS. In addition to all this, theres a whole lot of configuration which needs to be done to setup a Phoenix app for deployment, and it's hard to remember.

For that reason, I wanted to use Docker so that all of my deployments would be automated and reproducable. In addition, Docker would allow me to have reproducable builds for my releases. I could build my releases on any machine that I wanted in a contai

adamcichy

extension CGImage {
    var isDark: Bool {
        get {
            guard let imageData = self.dataProvider?.data else { return false }
            guard let ptr = CFDataGetBytePtr(imageData) else { return false }
            let length = CFDataGetLength(imageData)
            let threshold = Int(Double(self.width * self.height) * 0.45)
            var darkPixels = 0
            for i in stride(from: 0, to: length, by: 4) {
                let r = ptr[i]