A curated guide for the next generation of builders
When my nephew asked me how to learn computer science today, I realized the answer has fundamentally changed. Claude can write code. Copilot autocompletes functions. The skill isn't typing syntax anymore—it's understanding systems deeply enough to direct these tools and know when they're wrong.
This guide isn't about becoming a human compiler. It's about building the mental models that let you leverage AI as a force multiplier while understanding what's actually happening under the hood.
┌─────────────────────────────────────┐
│ Build & Ship Real Things │ ← Where value is created
├─────────────────────────────────────┤
│ Understand AI/ML Fundamentals │ ← The new literacy
├─────────────────────────────────────┤
│ Systems Thinking (OS, DB, Net) │ ← Why things work
├─────────────────────────────────────┤
│ Math (Linear Algebra, Calculus) │ ← The language of ML
├─────────────────────────────────────┤
│ Programming Fundamentals │ ← Still the foundation
└─────────────────────────────────────┘
You still need to understand code—not to write every line, but to read, debug, and direct AI effectively.
- Beginning C++ Programming (Udemy) — Understanding memory, pointers, and low-level concepts makes you a better programmer in any language
- Python — The lingua franca of AI/ML. Pick it up through projects.
You can't understand transformers if linear algebra is a black box.
-
3Blue1Brown: Essence of Linear Algebra — The most beautiful math videos on the internet. Watch this before touching any ML course. Builds geometric intuition for vectors, matrices, eigenvalues.
-
3Blue1Brown: Essence of Calculus — Derivatives and integrals visualized. Essential for understanding backpropagation.
-
Gil Strang's Introduction to Calculus (MIT OCW) — A legendary professor making calculus accessible.
-
3Blue1Brown: Neural Networks — What is a neural network actually doing? Visual intuition before diving into code.
AI can write CRUD apps. Understanding why systems are designed the way they are is what separates engineers from prompt-typers.
- CS162: Operating Systems (Berkeley) — Processes, threads, memory, file systems. When your ML model is slow, you'll know where to look.
- CMU Intro to Database Systems — Andy Pavlo is an incredible lecturer. Understand how data is stored, indexed, queried.
- Python API Development with FastAPI (freeCodeCamp) — APIs, HTTPS, SSL, SQL, database migrations. The full stack from scratch.
This is the new literacy. Even if you don't become an ML engineer, understanding how these systems work is non-negotiable.
Andrej Karpathy (ex-Tesla AI Director, OpenAI founding member) created the best "learn by building" ML course that exists.
- Neural Networks: Zero to Hero (Full Playlist) — Build neural networks from scratch in Python. No frameworks, no magic.
Key videos from this series:
| Video | What You'll Build |
|---|---|
| The spelled-out intro to neural networks | Micrograd - backprop from scratch |
| Building makemore | Bigram language model |
| Building makemore Part 2 | MLP language model |
| Building makemore Part 3 | BatchNorm, deeper networks |
| Building makemore Part 4 | Becoming a backprop ninja |
| Building makemore Part 5 | WaveNet architecture |
| Let's build GPT | GPT from scratch - the crown jewel |
| Let's reproduce GPT-2 | Production-grade implementation |
- Intro to Large Language Models (1hr overview) — Karpathy's accessible overview of how LLMs work, their capabilities, and limitations.
-
Deep Learning Specialization (Coursera/DeepLearning.AI) — Andrew Ng's comprehensive course. More structured than Karpathy, good for fundamentals.
-
Neural Networks and Deep Learning (Free Book) — Michael Nielsen's beautifully written online book. Great companion to video courses.
-
Fast.ai Practical Deep Learning — Top-down approach. Build things first, understand theory later. Jeremy Howard is an excellent teacher.
-
3Blue1Brown: But what is a GPT? (Visual intro to transformers) — Transformers visualized beautifully
-
3Blue1Brown: Attention in transformers, visually explained — The attention mechanism demystified
-
The Illustrated Transformer (Blog) — Jay Alammar's legendary visual explanation
Theory without practice is useless. This is where compounding begins.
- CLI tool — Build something you'll actually use
- API + Database — CRUD app with authentication
- Fine-tune a model — Take a small LLM, fine-tune on your data
- RAG application — Build a chatbot over your own documents
- Agent — Build an AI agent that can use tools
- Contribute to open source — Find an ML project, fix a bug, add a feature
The field moves fast. These channels help you keep up:
- Two Minute Papers — Digestible summaries of cutting-edge research
- Yannic Kilcher — Deep dives into important papers
- AI Explained — Thoughtful analysis of AI developments
- Lex Fridman Podcast — Long-form conversations with researchers (Karpathy, Sutskever, Altman, etc.)
- Latent Space — Technical AI podcast for practitioners
This is the paradox: you're learning CS while AI can do CS. Here's the framework:
- Use AI to explain concepts you don't understand
- Use AI to debug errors after you've tried yourself
- Use AI to explore rabbit holes — ask "why?" repeatedly
- Use AI to generate practice problems
- Take notes on what you learn from AI sessions (or it's wasted)
- Don't copy-paste AI code without understanding it
- Don't skip the struggle — confusion is where learning happens
- Don't use AI to complete assignments without doing them yourself first
Think of AI as a brilliant tutor who's available 24/7, not as a replacement for your brain. The goal is to build mental models that let you:
- Know what to ask for
- Evaluate if the answer is correct
- Debug when things go wrong
- Architect systems that don't exist yet
From engineers who've walked this path:
"Pick a niche and go deep." Put all your chips on one thing initially (fullstack, ML, systems). Breadth comes later. Depth compounds.
"Project-based learning has compounding benefits." After each project, up the difficulty. You'll learn faster each time.
"Do internships while in college." Don't wait for the perfect moment. Real-world experience beats classroom learning.
"Hackathons are great, but optimize for learning, not prizes." Build cool things. Winning is a side effect.
"Use AI to explore, but take notes." Retaining information from AI sessions requires active effort. Write things down.
| Month | Focus | Key Outcome |
|---|---|---|
| 1 | Programming fundamentals + Linear Algebra | Can read/write basic code, understand vectors/matrices |
| 2 | Calculus + Systems (OS basics) | Understand derivatives, processes, memory |
| 3 | Databases + First ML concepts | Can design schemas, understand what a neural net does |
| 4 | Karpathy's Zero to Hero | Built neural nets from scratch, understand backprop |
| 5 | Transformers + Build GPT | Understand modern LLMs, built one yourself |
| 6 | Ship a real project | Something deployed that solves a real problem |
- 3Blue1Brown: Linear Algebra
- 3Blue1Brown: Calculus
- 3Blue1Brown: Neural Networks
- Gil Strang's Calculus (MIT)
- C++ Programming (Udemy)
- CS162: Operating Systems (Berkeley)
- CMU Database Systems
- Python API Development (freeCodeCamp)
- Karpathy: Neural Networks Zero to Hero
- Karpathy: Let's build GPT
- Karpathy: Intro to LLMs
- Deep Learning Specialization (Coursera)
- Neural Networks and Deep Learning (Book)
- Fast.ai Course
This guide was compiled from recommendations by engineers at VideoDB. Last updated: March 2025.
The best time to start was yesterday. The second best time is now.