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.

koshatul

gpg-agent setup

Need to setup gpg-agent first, on OSX I use keychain (it also does ssh-agent)

$ brew info keychain
keychain: stable 2.8.5
User-friendly front-end to ssh-agent(1)
https://www.funtoo.org/Keychain
/usr/local/Cellar/keychain/2.8.5 (7 files, 108.5KB) *

mnylen

Simple implementation of debounced fetching in GraphQL to allow merging of multiple rest / database requests into one. Although this example uses GraphQL, the debouncedFetch / fetchProgramPlaycount implementations could probably be used in any context to achieve the same result.

This approach was first described by @leebyron at graphql/graphql-js#19 (comment)

For example this allows turning ten requests for playcounts from this GraphQL query into just one:

{
  latestPrograms(first: 10) {
    name,

playcount

kmpm

/// Original C++ implementation found at http://www.wilmott.com/messageview.cfm?catid=10&threadid=38771
/// C# implementation found at http://weblogs.asp.net/esanchez/archive/2010/07/29/a-quick-and-dirty-implementation-of-excel-norminv-function-in-c.aspx
/*
 *     Compute the quantile function for the normal distribution.
 *
 *     For small to moderate probabilities, algorithm referenced
 *     below is used to obtain an initial approximation which is
 *     polished with a final Newton step.
 *
 *     For very large arguments, an algorithm of Wichura is used.