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.

justjavac

// 需要通过 node --allow-natives-syntax 参数运行
const { GCSignal, consumeSignals, trackGarbageCollection } = require("gc-signals");

function fn() {
    // 创建两个局部变量
    const a1 = new GCSignal(1);
    const a2 = new GCSignal(2);
    global.tmp = a2;  // 将 a2 变量赋值给全局变量
    const o = {};
    trackGarbageCollection(o, 3); // 跟踪局部变量 o 的 GC 状态，标识为 3

justjavac

%GetOptimizationStatus return a set of bitwise flags instead of a single value, to access the value, you need to take the binary representation of the returned value. Now, for example, if 65 is returned, the binary representation is the following:

(65).toString(2).padStart(12, '0');
// 000001000001

Each binary digit acts as a boolean with the following meaning:

justjavac

// JavaScript 字符串编码使用 UTF-16
"💩".length === 2;
"💩" === "\u{1F4A9}";    // es6
"💩" === "\uD83D\uDCA9"; // es5

// 同一个编码可能使用不同的码位
"ò" === "ò";       // ❎
"ò" === "\xF2";    // ✅
"ò" === "o\u0300"; // ✅
"o\u0300".normalize("NFC") === "\xF2"; // ✅ es6 提供了 normalize 函数

bmeurer

"use strict";

// These are examples how to created PACKED_*_ELEMENTS arrays preinitialized
// in V8, following up on offline discussion at the last MunichJS meetup.

function createPackedViaArrayFrom(length, value) {
  return Array.from.call(null, Array.prototype.map.call({length}, _ => value));
}

function createPackedViaGenerator(length, value) {

acdlite

Demo: Coordinating async React with non-React views

tl;dr I built a demo illustrating what it might look like to add async rendering to Facebook's commenting interface, while ensuring it appears on the screen simultaneous to the server-rendered story.

A key benefit of async rendering is that large updates don't block the main thread; instead, the work is spread out and performed during idle periods using cooperative scheduling.

But once you make something async, you introduce the possibility that things may appear on the screen at separate times. Especially when you're dealing with multiple UI frameworks, as is often the case at Facebook.

How do we solve this with React?

justjavac

function test() {
  const a = new Int32Array(10);
  for (let i = 0; i < 1e6; i++) {
    a.set([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]);
  }
  return a;
}

console.time('test');
test();

justjavac

"use strict";

class Point {
  getX() {
    console.log(this.x); // C
  }
}

class ColorPoint extends Point {
  constructor() {

gaearon

// ------------
// counterStore.js
// ------------

import {
  INCREMENT_COUNTER,
  DECREMENT_COUNTER
} from '../constants/ActionTypes';

const initialState = { counter: 0 };

justjavac

var cv = document.getElementById('cv');
var c = cv.getContext('2d');
var txtDiv = document.getElementById('txt');
var fileBtn = document.getElementById("up-button");
var img = new Image();
img.src = 'a.jpg';
img.onload = init; // 图片加载完开始转换
fileBtn.onchange = getImg;

// 根据灰度生成相应字符