🦸 AI Hero

A complete, interactive course on AI. Start with a single neuron, end with your own GPT. Math, intuition, theory, and hands-on construction — at your own pace.

Open the [[course/index|🎓 Course]] first. Full-screen Liquid Glass page with all videos embedded (not just linked), every stage's playlist playing inline, and working clickable navigation to the dashboard + 5 visualization webapps + each module page.

You will cover the math, develop intuition for the architecture, and build an LLM from the ground up — starting with a perceptron in pure Python, finishing with your own GPT in PyTorch, and closing with a research paper you can read critically.

How long will this take?

~20 hours of focused work. Plan ahead — do it in small chunks so things sink in. Interactive webapps, embedded videos, and live notebooks make every concept something you can poke at, not just read about.

The 5 Stages

#	Stage	Video	Exercise	Total
1	🧮 Math Prerequisites	4 h	—	4 h
2	🧠 Intuition	3 h	—	3 h
3	📜 Academic Theory	1 h	1 h	2 h
4	🏗️ Hands-On (Karpathy)	12 h	5 h	17 h
5	🔍 Research Paper	—	1 h	1 h

20 h is the video-at-1x estimate. 2x is often fine. Exercise time is the floor — go deeper if you have it.

Start here

1. Read [[00-getting-started/index|How to use this cabinet]] first — it explains the agents, the notebooks, and the progress system.
2. Open the Dashboard (left sidebar) to see your progress at a glance.
3. Work through the stages in order. Each stage links to the next.

The stages

• 🧮 [[01-math-prerequisites/index|1. Math Prerequisites]] — linear algebra, calculus, probability (4 h)
• 🧠 [[02-intuition/index|2. Intuition]] — 3Blue1Brown's neural network series (3 h)
• 📜 [[03-academic-theory/index|3. Academic Theory]] — survey of modern architectures (2 h)
• 🏗️ [[04-hands-on/index|4. Hands-On]] — Karpathy's Zero to Hero, from micrograd to GPT (17 h)
• 🔍 [[05-research-paper/index|5. Research Paper]] — read one paper critically (1 h)

🎛️ Interactive visualizations

Five live webapps, cabinet parchment theme, editorial typography. Play with them while you watch the videos.

• 📊 [[dashboard/index|Dashboard]] — your progress across all stages, with per-section bars
• 🗺️ [[concept-map/index|Concept Map]] — force-directed graph of 38 concepts, click any node to locate it in the course
• 🧠 [[neuron-playground/index|Neuron Playground]] — drag sliders for inputs + weights, watch a single neuron compute live with its activation curve
• 🎢 [[gradient-descent/index|Gradient Descent]] — drop a ball on a 2D loss landscape, tune LR and momentum, break it on purpose
• 👁️ [[attention-viz/index|Attention Visualizer]] — click tokens in a sentence, see the attention heatmap light up, toggle the causal mask

Your team

Two agents live in this cabinet — ask either anything. They are not ChatGPT: do not ask them to do your exercises for you. They are study partners, not spoiler machines.

• 🎓 Mentor — plans your week, sets pace, flags when you're behind, explains math intuitions (matrix ops, chain rule, probability), and owns progress/progress.csv + the dashboard
• 💻 Code Tutor — reviews your notebook attempts; explains PyTorch + Python patterns

Ground rules

1. Do exercises yourself. Wrong attempts teach more than copying a solution.
2. Type the code. Do not copy-paste from Karpathy's videos. Your fingers need the reps.
3. Write in the notes fields. Each section has a "My notes" block. Future-you needs them.
4. If you're stuck > 30 min, ask the tutors — don't ask ChatGPT (it'll give you spoilers).

When you finish

You should be able to:

• Write forward + backward pass of a small MLP by hand on paper
• Explain attention as a weighted-sum-over-values in one sentence
• Train a transformer from scratch on a toy problem
• Read an arXiv abstract and tell whether it's worth your time

Good luck. Ship it. 🚀