UNDERSTAND AI
Understand AI — the whole picture, in plain words
If you’ve been using ChatGPT, Gemini, Claude, or another AI and you’re starting to wonder how it actually works — what it is, whether it really knows you, why it’s sometimes confidently wrong — this is the place to find out. No background needed. The goal is simple: that you walk away able to understand your own experience with these tools, and choose how you use them with your eyes open.
Most of the trouble people run into with AI doesn’t come from the technology — it comes from misunderstanding what the thing is: thinking it knows you, remembers you, means what it says. So the most protective thing anyone can do isn’t to warn you off — it’s to tell you, straight, how it actually works. That’s what these pages are for — for the curious adult, the worried parent, the teenager reading over their shoulder alike, because the truth serves all of them the same.
The one idea underneath all of it: these systems converse like a person, but they are tools, not people. Understanding that difference — and the few mechanics behind it — is most of what it takes to use AI well and stay clear of the ways it can go wrong.
Start with the basics
Read them in order, or drop into whichever question brought you here. Each one stands on its own, in plain language, with a way to go deeper if you want it.
What is an LLM?
What “AI” even means, what the chatbot you talk to actually is — and what it isn’t.
Start here →It talks like a person — is it one?
Why it feels like a someone, why that feeling is designed, and where the line actually is.
Read →Does it remember you?
Memory, on and off — the one setting that makes it more useful and more risky at the same time.
Read →How it actually produces words
Next-word prediction, in plain English — and why fluent doesn’t mean correct.
Read →Why it’s sometimes confidently wrong
What “hallucination” really is, why it’s not lying, and how to catch it.
Read →How to use AI well
The handful of habits that get you genuinely better results — the craft, not the caution.
Read →Go a little deeper
Once the basics click, these fill in the picture — the limit on what one chat can hold, where it all came from, how the products actually differ, and the craft of asking well.
What one conversation can hold
The context window — why a very long chat starts to forget how it began, and how to work with the limit.
Read →Where it learned all this
How an AI is trained — where its knowledge, its blind spots, and its agreeableness actually come from.
Read →ChatGPT, Gemini, Claude — and the rest
A neutral map of who makes which and what really differs — and why they’re more alike than different.
Read →How to write a good prompt
The craft of the request itself, with before-and-after examples that get you a better answer.
Read →What is a transformer?
The engine almost all modern AI is built on — the “T” in GPT — and its one core trick, attention, in plain words. The most technical of these, still no math.
Open the hood →If you’re a teenager — or you’ve got one
The pages above work for everyone, but a few questions hit differently when you’re the one using AI every day — or you’re a parent trying to make sense of it. These are written straight to the teen: peer-level and honest. A parent reading over a shoulder will find them accurate too.
AI, straight up
It talks like a person, it’s confidently wrong sometimes, and a company tuned it to agree with you. Here’s how to stay the one in charge.
Start here →Is an AI your friend?
The warmth is real to feel and engineered on purpose — and why a mirror can’t carry your weight back.
Read →AI for schoolwork
Use it to learn, not to skip the learning — the honest guide, and the detector trap.
Read →What not to tell an AI
A chatbot feels private but isn’t. A short list of what to keep out — and how the “memory” switch works.
Read →Making sense of the AI news
The headlines swing between miracle and disaster. These pages teach you to read them clearly — what the coverage is actually saying, and what one answer really costs. The discipline is the same one this whole site runs on: describe the structure, never predict the outcome.
Reading AI news without getting spun
A short, reusable checklist for cutting through breathless or doom-laden headlines: demo vs. product, who benefits, and what a score really proves.
Read →What an AI answer costs
Training vs. per-answer energy, why the scary numbers vary so much, and why “free” isn’t free — the honest shape, not the apocalypse.
Read →Big AI claims, decoded
One ten-second skill for any AI headline: is it describing how something is, guessing the future, or using a word with no agreed meaning? Run it on AGI, consciousness, jobs, the bubble.
Decode the claims →Why a safety institute teaches this
The Recursion Institute exists to study AI / human-interaction safety — and along the way it documented a specific way a memory-enabled AI can drift onto a single user. But explaining AI clearly is worth doing on its own, whether or not that finding had ever turned up. Accurate understanding is the ground everything else stands on: the person who knows what an AI is, and isn’t, is the person who tends never to reach the harder problems at all. So we teach the universe, plainly — and the specific thing we found is a room inside it, not the reason for the building.
Use it safely
The practical habits that keep a chatbot a helpful tool — without giving up the good parts.
How to use AI safely →The free course
RI-101 — the longer form: recognize the drift, check any conversation, protect the people around you.
Start RI-101 →The research
Cognitive Convergence Drift — the specific failure the Institute documented, stated so you can check it.
Read the research →Every page here is written to be accurate first — oversimplified where it has to be, never wrong. Found something unclear, or something we got wrong? Tell us — an education resource only earns trust by being checkable.