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.

iTrooz

// Automatically use pino-pretty if available in Cloudflare Workers 
// Note that pino-pretty needs to be installed in prod too with this solution, but it will not be used
// https://gist.github.com/iTrooz/179e498d181f811ec6ebc30396a73f51
import { pino, type DestinationStream } from "pino";
import { Writable } from "stream";
import pinoPretty from "pino-pretty";

// Setup pino logger, using pino-pretty if available (should only be used for dev, not prod !)
export function setupLogger() {
  // See later

takunagai

Tailwind CSS v3 to v4 Migration Guide

This document provides a comprehensive guide to help you understand and migrate from Tailwind CSS v3 to v4. It focuses on code-level changes with clear examples showing both v3 and v4 equivalents.

Browser Compatibility Note: Tailwind CSS v4.0 is designed for Safari 16.4+, Chrome 111+, and Firefox 128+. If you need to support older browsers, stick with v3.4 until your browser support requirements change.

Installation Changes

Simplified Installation (v3 vs v4)

erkobridee

/*
  useful references:
  
  https://github.com/erkobridee/lab-pdf-lib/blob/83ef7e53e61227cef4293a6294979b1e551a3603/src/helpers/random.js#L4
  
  https://github.com/erkobridee/nx-todos/blob/1647f66518a6def9ddf7569ca7a2b6397337d09c/libs/shared/helpers/src/lib/values.ts#L25
  
  https://github.com/erkobridee/gatsby-typescript-app-starter/blob/5a7bcfbbe6ff014ecb360da35a3e8255fe80bf9e/src/helpers/values.ts#L21
  
  https://gist.github.com/rubinchyk/46d9868e2eb86d744abcb83b5b6d8d29

erkobridee

Using multiple git accounts on the same machine

• Managing multiple accounts | GitHub Docs

  • Contributing to multiple accounts using SSH and GIT_SSH_COMMAND

GIT_SSH_COMMAND='ssh -i PATH/TO/KEY/FILE -o IdentitiesOnly=yes' git clone git@github.com:OWNER/REPOSITORY

felixeichler

import { useMouse } from '@uidotdev/usehooks';
import * as React from 'react';

interface Positions {
    /* Sub-menu height */
    h: number;
    /* Mouse x */
    mouseX: number;
    /* Mouse y */
    mouseY: number;

erkobridee

• [StackOverflow] How do I get info about a Youtube video's chapters from the API?

  • R: using a provided open-source API

    • [GitHub] Benjamin-Loison / YouTube-operational-API - site: YouTube operational API | API call - get video chapters info example

  • R: using javascript function at the youtube video page

erkobridee

/*
    from:

    Enums considered harmful | Matt Pocock
    https://www.youtube.com/watch?v=jjMbPt_H3RQ
*/

const LogLevel = {
  DEBUG: "DEBUG",
  INFO: "INFO",

erkobridee

"Any sufficiently advanced technology is indistinguishable from magic." by - Arthur C. Clarke | CCCB LAB

Arthur C. Clarke was a scientist, science writer and author of science fiction stories and novels. This quote of his reflects on the early beginnings of technology and offers us a way of linking up the two exhibitions that have been held this year at the CCCB: Black Light and Stanley Kubrick.

this quote it's also refered on [YouTube] Why Magic Systems don't feel Magical | Tale Foundry - 2023/05/31

Learning

• 🇧🇷 [YouTube] Rotina de estudos: a melhor estratégia da neurociência| Gustavo Kobayashi - 2022/10/10

steven-tey

// Note: this gist is a part of this OSS project that I'm currently working on: https://github.com/steven-tey/dub

export default async function getTitleFromUrl (url: string) {
  const controller = new AbortController();
  const timeoutId = setTimeout(() => controller.abort(), 2000); // timeout if it takes longer than 2 seconds
  const title = await fetch(url, { signal: controller.signal })
    .then((res) => {
      clearTimeout(timeoutId);
      return res.text();
    })