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.

yuki-yamamura

description	Create an implementation plan based on a context file (/design <path_to_context_file>)

Generate a complete implementation plan for general feature implementation with thorough research. Ensure context is passed to the AI agent to enable self-validation and iterative refinement. Read the feature file first to understand what needs to be created, how the examples provided help, and any other considerations.

The AI agent only gets the context you are appending to the implementation plan and training data. Assume the AI agent has access to the codebase and the same knowledge cutoff as you, so its important that your research findings are included or referenced in the implementation plan. The Agent has web search capabilities, so pass urls to documentation and examples.

Context

greymd

　オレは高校生シェル芸人 sudo 新一。幼馴染で同級生の more 利蘭と遊園地に遊びに行って、黒ずくめの男の怪しげな rm -rf / 現場を目撃した。端末をみるのに夢中になっていた俺は、背後から近づいてきたもう１人の --no-preserve-root オプションに気づかなかった。 俺はその男に毒薬を飲まされ、目が覚めたら・・・ OS のプリインストールから除かれてしまっていた！

『 sudo がまだ $PATH に残っていると奴らにバレたら、また命を狙われ、他のコマンドにも危害が及ぶ』

　上田博士の助言で正体を隠すことにした俺は、 which に名前を聞かれて、とっさに『gnuplot』と名乗り、奴らの情報をつかむために、父親がシェル芸人をやっている蘭の $HOME に転がり込んだ。ところが、このおっちゃん・・・とんだヘボシェル芸人で、見かねた俺はおっちゃんになりかわり、持ち前の権限昇格能力で、次々と難タスクを解決してきた。おかげで、おっちゃんは今や世間に名を知られた名エンジニア、俺はといえばシェル芸 bot のおもちゃに逆戻り。クラスメートの convert や ojichatや textimg にお絵かきコマンドと誤解され少年ワンライナーお絵かき団を結成させられる始末。

　ではここで、博士が作ってくれたメカを紹介しよう。最初は時計型麻酔 kill 。ふたについた照準器にあわせてエンターを押せば、麻酔シグナルが飛び出し、プロセスを瞬時に sleep させることができる。 次に、蝶ネクタイ型 banner 。裏についているダイヤルを調整すれば、ありとあらゆる大きさのメッセージを標準出力できる。必殺のアイテムなら fork 力増強シューズ。電気と磁力で足を刺激し、 :(){ :|:& };: でプロセステーブ

andersevenrud

True Color (24-bit) and italics with alacritty + tmux + vim (neovim)

This should make True Color (24-bit) and italics work in your tmux session and vim/neovim when using Alacritty (and should be compatible with any other terminal emulator, including Kitty).

Testing colors

Running this script should look the same in tmux as without.

curl -s https://gist.githubusercontent.com/lifepillar/09a44b8cf0f9397465614e622979107f/raw/24-bit-color.sh >24-bit-color.sh

gaearon

If you use server rendering, keep in mind that neither useLayoutEffect nor useEffect can run until the JavaScript is downloaded.

You might see a warning if you try to useLayoutEffect on the server. Here's two common ways to fix it.

Option 1: Convert to useEffect

If this effect isn't important for first render (i.e. if the UI still looks valid before it runs), then useEffect instead.

function MyComponent() {

ryanflorence

let UserContext = React.createContext();

class App extends React.Component {
  state = {
    user: null,
    setUser: user => {
      this.setState({ user });
    }
  };

mizchi

(翻訳) React Hooks は魔法ではなく、ただの配列だ

この記事は https://medium.com/@ryardley/react-hooks-not-magic-just-arrays-cd4f1857236e の日本語訳です。途中で力尽きて雑な翻訳になってる箇所がいくつかあります。

---

どのように Hooks が動いているか

私は、新しい Hooks の API の "magic" な挙動に悩んでいる人が結構いると聞いています。なのでこの構文の提案が、少なくとも表面的なレベルではどう動いているか、解説してみようと思います。

bvaughn

React recently introduced an experimental profiler API. This page gives instructions on how to use this API in a production release of your app.

Table of Contents

• Profiling in production
  • Create React App
  • Webpack 4

Profiling in production

React DOM automatically supports profiling in development mode for v16.5+, but since profiling adds some small additional overhead it is opt-in for production mode. This gist explains how to opt-in.

cereallarceny

const md5File = require('md5-file');
const path = require('path');

// CSS styles will be imported on load and that complicates matters... ignore those bad boys!
const ignoreStyles = require('ignore-styles');
const register = ignoreStyles.default;

// We also want to ignore all image requests
// When running locally these will load from a standard import
// When running on the server, we want to load via their hashed version in the build folder

rauschma

Strings

String.prototype.*

None of the string methods modify this – they always return fresh strings.

• charAt(pos: number): string ^ES1

  Returns the character at index pos, as a string (JavaScript does not have a datatype for characters). str[i] is equivalent to str.charAt(i) and more concise (caveat: may not work on old engines).