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.

zmagg

How much prep did you do for the technical interviews?

I get asked this a lot so I thought I'd write it up to share. For context, I was interviewing for staff+ individual contributor roles at tech companies.

Technical Interviews

I was preparing to do some amount of live coding interviews, and some amount of pairing interviews.

Tech setup check-in

My friend Tom offered to do a pairing interview with me, which I also used to check my Zoom / screenshare setup. I screen shared from my 2019 XPS 13 for the pairing, but also had Zoom on an ipad for video/audio. I used Airpods for audio. This setup actually worked great, but it was nice to dial in some of the specifics--mic feedback from dialing in from two places, and just practice my schpiel about explaining what I was doing to the other person. Probably not useful for you unless you too, have Linux problems.

IanColdwater

Mute these words in your settings here: https://twitter.com/settings/muted_keywords

ActivityTweet
generic_activity_highlights
generic_activity_momentsbreaking
RankedOrganicTweet
suggest_activity
suggest_activity_feed
suggest_activity_highlights
suggest_activity_tweet

just-be-dev

#!/bin/bash

DAYS_SINCE_LAST_CHANGE=14   # If a project hasn't been touched in this long its node_modules will be deleted
SEARCH_PATH="./Git"         # Update this to the path where your code is stored

TOTAL_BYTES_REMOVED=0
Mb=1000000
Kb=1000

node_modules=$(find $SEARCH_PATH -name "node_modules" -type d -prune)

jvcleave

Nightly required - get link from the bottom of this page (e.g. https://openframeworks.cc/ci_server/versions/nightly/of_v20190324_linuxarmv7l_nightly.tar.gz)
https://openframeworks.cc/download/


Download OF and unpack:

wget https://openframeworks.cc/ci_server/versions/nightly/of_v20190324_linuxarmv7l_nightly.tar.gz
tar -zxvf of_v20190324_linuxarmv7l_nightly.tar.gz
mv of_v20190324_linuxarmv7l_nightly openFrameworks

scottrigby

#!/bin/sh

help() {
    cat << EOF
spin(1)

NAME
    spin - begin or end a spinner background process

SYNOPSIS

beesandbombs

int[][] result;
float t, c;

float ease(float p) {
  return 3*p*p - 2*p*p*p;
}

float ease(float p, float g) {
  if (p < 0.5) 
    return 0.5 * pow(2*p, g);

beesandbombs

// zooming squares. by dave

int[][] result;
float t, c;

float ease(float p) {
  return 3*p*p - 2*p*p*p;
}

float ease(float p, float g) {

jstepien

[package]
name = "bk"
version = "0.1.0"
authors = ["Jan Stępień"]

karpathy

"""
Minimal character-level Vanilla RNN model. Written by Andrej Karpathy (@karpathy)
BSD License
"""
import numpy as np

# data I/O
data = open('input.txt', 'r').read() # should be simple plain text file
chars = list(set(data))
data_size, vocab_size = len(data), len(chars)