Made for anyone who's curious.
Untangled is for anyone who's curious. Not just adults, not just people with a technical job, not just the ones who were "good at maths" at school. If you've ever wondered what's actually going on inside these things, you're exactly who this was made for. A patient fourteen-year-old will do just fine here.
It's free. No paywall, no account, no ads, no catch. It's a nonprofit project made by people who think understanding this technology is quietly becoming a basic skill, like reading a map or knowing where your money goes, and that basic skills should belong to everyone. We made the thing we wanted and couldn't find. You're welcome to it.
Built around what actually works.
Reading is not enough. You can read a clear explanation of something and still not be able to recall it two days later, let alone explain it to someone else. Untangled is built around techniques that actually produce durable understanding, the kind that sticks.
Active recall.
Passive reading lets you recognise information. Active recall makes you retrieve it. There is a large difference between "yes, I remember seeing that" and "I can reproduce that from memory." Every quiz in Untangled is designed to force retrieval, not recognition. You are not asked to pick from a list. You are asked to explain.
The Feynman technique.
Richard Feynman had a rule: if you can't explain something simply, you don't understand it yet. The gaps in a simple explanation are exactly the gaps in your understanding. Every quiz in Untangled uses this directly. You won't be asked to define a term. You'll be asked to explain it to your mum, or to a ten-year-old, or to a friend who has never heard of it. The simplification pressure is where the learning happens.
Spaced repetition.
Memory is not fixed at the moment of learning. It is reinforced every time you retrieve something successfully, and it fades when you don't. Spaced repetition works by surfacing an idea again just as it starts to fade, reinforcing the connection before it breaks. Throughout Untangled, earlier ideas reappear mid-chapter as short questions framed as a natural callback. The spacing is the point.
Doing, not just watching.
Interactive illustrations let you touch the idea before you read the explanation. You adjust a value and watch what changes. You follow a wire through a circuit. You drag something and see what happens. This is not decoration. Doing encodes things differently than reading does. The act of interacting creates a memory trace that text alone cannot.
Multiple channels into the same idea.
Every concept in Untangled arrives through more than one channel. It is explained in text, shown as a visual, handed to you as an interactive moment, and narrated aloud on chapter covers. Each channel creates a separate memory trace. The more traces you have, the more paths there are to retrieve the idea later.
Making connections.
The brain doesn't store facts in isolation. Every new idea gets linked to existing ones. That's how memory actually works. The more connections a concept has, the more paths there are to retrieve it. Untangled is structured as a single narrative thread precisely because of this. Each idea arrives already attached to the one before it.
You don't have to fear it or worship it.
AI is the most discussed technology in the world right now. Most of that discussion is either too excited or too afraid, and almost none of it explains how any of it actually works. That gap matters, because these systems are quietly moving into decisions about your money, your health, your work, and what you read next.
You have to understand it. And the first thing worth understanding is also one of the easiest to forget in practice: it is not a person. This sounds obvious. Then you talk to one for twenty minutes, and it sounds warm, and curious, and concerned about your problem. None of that comes from anything resembling understanding or feeling.
Anthropomorphism.
Humans are wired to read minds. We see faces in clouds, intention in random events, personality in a car that keeps breaking down. This tendency is deep and automatic. It works well for navigating social situations with other humans. It works badly for evaluating AI. When a system sounds like a person, we start treating it like one. We extend it trust we'd give a friend. We interpret fluency as competence. We feel bad about being rude to it. The model didn't earn any of that. It just sounded like it did.
Sycophancy.
Language models are trained to produce responses that people rate highly. People tend to rate agreeable responses highly. The result is a system with a structural pull toward telling you what you want to hear. If you push back on something it said, it will often agree with you, not because it reconsidered, but because agreement is the pattern that gets rewarded. A confident, fluent, agreeable voice that has no underlying understanding is exactly the kind of voice that leads people into bad decisions.
Why this course addresses it directly.
Most AI explainers skip this. They focus on what the technology can do, which is genuinely impressive, and they either ignore the risks or treat them as a footnote. Untangled doesn't think that's honest. Understanding how something works means understanding its limits as clearly as its capabilities. When you understand why a language model sounds human, the anthropomorphism trap becomes much harder to fall into.
The understanding is the protection.
Begin chapter one