Chapter 00 · before you begin

Start here!

Just a few quick notes on who I am, how these notebooks work, and prereqs.

~ 4 sectionsprereq: n/athen: ch.01

Introduction

Hello! My name is Conrad and I'm an undergraduate at Emory University studying Computer Science! My main research interests are in mechanistic interpretability and evaluation awareness.

Since 9th grade in high school, I was a diehard premed kid. But recently, I got the bright idea to pivot to CS (in this job market??), and more specifically, machine learning.

To me, ML is kinda like a hibachi chef doing the egg trick: the adults (ML researchers) instantly see what's going on, and most kids just gape in awe. I'm the one squinting, trying to catch the trick.

You only catch the trick by slowing the chef down.

That's how learning ML worked for me—frame by frame, every piece broken apart until it clicked. And learning it that slowly means I tend to recognize where people get stuck, because I got stuck there too.

How many eggs hopped into the hibachi hat?

Oh… just broken eggs

fig. 00 — looking inside the hat

So for the curious minds that want to become hibachi egg-trick masters? This place is for you.

Please use AI (this site will help you too!)

Around late 2022, I started using AI to help me learn. In the early days, asking it a question was like shaking a magic 8-ball. But it's come far enough since then to be a genuine learning tool—so long as you use it correctly.

Asking AI "How does backpropagation work?" is one thing. Giving it context, telling it how to explain, and then asking is another entirely. The difference in what you get back is night and day:

the same question, two ways

your prompt

How does backpropagation work?

↓

what you get back

Backpropagation is an algorithm for training neural networks. It computes the gradient of the loss function with respect to each weight using the chain rule, propagating the error backward from the output layer through every hidden layer, so that an optimizer such as stochastic gradient descent can…

~generic — a wall of jargon you still have to decode

fig. 01 — i personally prefer Claude, but any chatbot works!

The catch is that the second takes more work, and that's exactly what this site is aims to solve. After considering the possibility of integrating a chatbot into my lessons, I realized that most people (especially CS people) already pay for am LLM of their choice, whether that's ChatGPT, Claude, Gemini, or something further off the beaten path. So here's how I structured things:

01
At the start of every lesson, click "Copy lesson primer into your own AI." Paste it in at the beginning of your chat session. It's a quick blurb that tells your AI what the lesson covers, how it's structured, and what to expect.
02
Whenever a sentence trips you up, highlight and copy it. You'll get a prompt containing info on where you are in the lesson and relevant context. That way, when you append your question, you'll get an answer that actually fits where you're stuck.
03
Never stop asking questions. Teachers often say that "there are no stupid questions." That's just not true... but AI doesn't care. With AI, ask the stupid, the obvious, and the embarrassing. All of them, as many times as it takes.

Prerequisites and helpful stuff

ML is often treated as gatekept territory, fenced off behind layers of math and STEM. However, when getting started, I think the prereqs are less than most people expect.

While these guides will focus heavily on the math, theory, and why behind ML, a decent understanding of the following topics will suffice:

Linear Algebra: For me, an introductory course in linear algebra was good enough. What helped even more though was 3Blue1Brown's Essence of Linear Algebra YouTube series.
Calculus: When I started ML, I only had background in AP Calculus BC. If you have this (as well as the will to learn a bit more calculus), then you should be fine for introductory ML. For those going even more advanced, matrix calculus is required.
Probability: To be completely honest, as of now I only have background up to AP Statistics. ML leans hard on probability theory, so this is one I'm actively picking up as I go. If anything, that makes me a decent guide through it: I'm hitting these gaps right alongside you, not narrating them from memory.
Python: I only knew basic Python when starting, and my best advice is to not let AI generate code for you without understanding. If you're relatively new to Python, a lot of the uphill battle will be syntax related. Ask those questions! It'll only help you.

▋

fig. 02 — you don't need to know all these buzzwords, but here's a preview!

Here's some gems that I've used to aid my understanding of ML.

3Blue1Brown: I could list many resources I've used while trying to learn, but I will insist say that Grant Sanderson's 3B1B occupies a tier no one else reaches. An entire generation (including me) learned to see linear algebra and calculus through his animations. He has an amazing series on ML as well as various other topics, please take the time to visit his channel.
MIT OCW: For those who are really interested in the math, MIT OpenCourseWare is an amazing resource. They have everything from multivariable to discrete math. Gilbert Strang's Linear Algebra course is legendary.
Anthropic's blogs: If you ever get burnt out, I suggest looking at Anthropic's Engineering page. These blogs are super interesting and have never failed to reignite my interests.
Lebron James: Shoutout just because.

Where to start

The first lesson is where it all begins. You'll build a two-layer network that reads handwritten digits, derive backpropagation yourself, and watch it learn — and everything after that builds on it.

next up

ch.01 — The fundamentals

The stuff behind the machine: learning forward pass to backpropagation using the math you already know.

Coming soon

Have fun!!