Free · No sign-up · At your own pace
Learn AI, from beginner to expert
Four levels, in order. Each mixes short guides, the key words, simulators, courses and a book, and ends with a quick self-check. Tick off steps as you go: your progress is saved in this browser. Short on time? Switch on "Essentials only" to skip the optional extras.
- Level 1FoundationsWhat AI is, what it can do and how to use it safely0/13 done
- Level 2How AI worksNeural networks, language models and how they learn0/12 done
- Level 3Tech and toolsThe building blocks: agents, RAG, APIs, chips and frameworks0/13 done
- Level 4Expert viewFrontier debates: scaling, AGI, safety and the rules0/13 done
Foundations
Understand what AI is, what it can and can't do, and how to use it safely.
By the end of this level you'll be able to
- Explain the difference between AI, machine learning, deep learning and generative AI
- Know what chatbots are good at, and where they go wrong
- Recognise the big milestones from 1950 to today
- Spot AI fakes and scams, and protect your privacy
- ReadWhere AI stands in 2026: the big picture in plain English
The big picture on one page: what AI can do today, who builds it and the big debates.
- ReadAI vs machine learning vs deep learning: how they fit together
The four circles everyone mixes up, with a demo you can try of how a machine learns.
- Try itOptionalTeachable Machine ↗
Train your own model from your webcam in five minutes, no code needed.
- ReadThe 5 best AI chatbots right now
Pick a chatbot to use, and learn what each of the big ones is best at.
- ReadThe AI timeline: key moments from 1950 to today
The key moments from Alan Turing to ChatGPT, and why each mattered.
- ReadOptionalAI fails: when AI got it badly wrong (and what to learn)
Real mistakes, from fake court cases to glue on pizza, and what they teach about AI's limits.
- ReadHow to spot AI fakes and scams: cloned voices, deepfakes and fake ads
Cloned voices, deepfakes and fake ads: how to spot them and stay safe.
- ReadUsing AI without worrying about your privacy
What happens to what you type, and simple settings that protect you.
- BookOptionalCo-Intelligence: Living and Working with AI ↗
A friendly, practical book on working alongside AI, if you want to read further.
- CourseOptionalElements of AI ↗
Prefer a structured course? The best free introduction to AI, with exercises and no maths.
- ReadOptionalThe best free AI courses, from beginner to expert
Nine free courses sorted by level, for when you want to go hands-on.
Check yourself
Three quick questions on level 1. No score is kept.
1. Which of these is the smallest circle, sitting inside all the others?
Generative AI is a kind of deep learning, which is a kind of machine learning, which is a kind of AI.
2. A chatbot confidently cites a court case that never existed. What is this called?
A hallucination is when an AI states something false as if it were true. Always check important facts.
3. A video call from your "boss" urgently asks you to transfer money. What's the safest move?
Voices and faces can be cloned. Checking through a separate channel you already trust is the reliable defence.
All right! You're ready for level 2: How AI works.
How AI works
Understand what happens inside a model: how it learns, how it reads your words and how it creates.
By the end of this level you'll be able to
- Explain how a neural network learns from examples
- Describe how a chatbot turns your words into an answer, token by token
- Name the stages of training: pre-training, fine-tuning and RLHF
- Explain how an image generator makes a picture from noise
- ReadHow AI chatbots work: from your question to the answer
The whole journey on one illustrated page: from your question, through tokens and attention, to the answer.
- WordsHow machines learn
The basic parts: networks, data, the settings they learn, and the main fields of AI.
- Try itOptionalTensorFlow Playground ↗
Watch a real neural network learn to separate dots, layer by layer.
- WordsInside a language model
Follow your question through a model: split into tokens, turned into numbers, weighed with attention, answered word by word.
- Try itOptionalDiffusion Explainer ↗
See an image appear out of random noise as a diffusion model works.
- ReadRAG explained: how AI looks things up before it answers
How AI looks things up before it answers, and why that cuts down on made-up answers.
- BookOptionalYou Look Like a Thing and I Love You ↗
A funny, illustrated book on how machine learning really works, and how it gets things wrong.
- CourseOptionalIntro to Machine Learning ↗
Build and test your first machine learning model in about three hours.
Check yourself
Three quick questions on level 2. No score is kept.
1. What does a large language model actually do when it writes an answer?
It predicts the most likely next piece of text again and again. That is also why it can sound sure while being wrong.
2. Which training stage teaches a model which answers people prefer, using their ratings?
RLHF (reinforcement learning from human feedback) uses people's rankings of answers to shape how the model responds.
3. How does a diffusion model create an image?
It learned to remove noise during training, and your prompt steers each step of the clean-up.
All right! You're ready for level 3: Tech and tools.
Tech and tools
Know the building blocks people use to make AI products, from APIs and agents to frameworks and chips.
By the end of this level you'll be able to
- Explain how apps use AI models through APIs, and what open-weight models are
- Understand RAG, vector databases and MCP
- Know what an AI agent is, how it uses tools and how it can be attacked
- Name the main frameworks and the hardware AI runs on
- ReadThe AI tech stack: the tools behind every AI product, explained
The whole map first: the eight layers behind every AI product, from chips to apps, and the names in each.
- WordsUsing models in products
How companies plug a model into their app, control how it behaves and test that it works.
- ReadOptionalHow to run AI on your own computer
Run an open model on your own laptop: private, offline and free.
- ReadAI agents, explained simply
From chatting to doing: how agents plan, use tools and check their work.
- WordsAgents and their tools
The standard that connects agents to apps, and the attack every builder worries about.
- Try itOptionalAgent Breaker ↗
Try to trick AI apps into breaking their own rules, and see why prompt injection is so hard to stop.
- WordsFrameworks and coding
PyTorch, TensorFlow, LangChain, Hugging Face, NVIDIA's CUDA, and building apps by describing them.
- ReadWhat powers AI: from chips to power plants
Follow one chatbot question down through chips, data centres and power plants.
- BookOptionalAI Engineering ↗
Chip Huyen's book on building real products with AI models, for when you want to go hands-on.
- CourseOptional5-Day AI Agents Intensive ↗
Google's free course on building AI agents, with hands-on coding labs.
Check yourself
Three quick questions on level 3. No score is kept.
1. A company's support chatbot answers from its own help pages, with links. Which technique is that?
RAG searches the documents first, often using a vector database, then answers from what it found.
2. What is MCP (Model Context Protocol) for?
MCP lets any AI app that supports it use any tool that offers an MCP server, like a USB-C port for AI.
3. PyTorch and TensorFlow are…
They handle the maths of neural networks. LangChain and LlamaIndex are frameworks for building apps on top of models.
All right! You're ready for level 4: Expert view.
Expert view
Follow the frontier: what drives progress, how close we are to human-level AI, and how AI is kept safe and governed.
By the end of this level you'll be able to
- Explain scaling laws and why labs are building huge data centres
- Discuss AGI using the levels the AI labs themselves use
- Understand alignment and the main approaches to AI safety
- Know how the EU AI Act sorts AI by risk, and how other countries differ
- ReadThe AI ladder: how close are we to really smart AI?
The five levels AI companies use to track progress towards AGI, and where today's models stand.
- ReadOptionalHow much energy does one AI question use?
What one AI question really costs in electricity and water, with sources.
- ReadOptionalAI data centres around the world: the good, the worries and the facts
The good, the worries and the facts about the buildings behind the AI race.
- WordsSafety and the rules
Keeping AI doing what people intend, and the first broad law that governs it.
- ReadEssays that shaped AI: 10 must-reads, from Turing to today
The ten essays people in AI keep quoting, from Turing to today, with optimists and sceptics side by side.
- ReadOptionalAI and cyber security: the new arms race
How attackers and defenders both use AI, in the new arms race.
- ReadOptionalAI predictions: the hits, the misses and the spooky ones
Who got the future of AI right and wrong, and why forecasting it is so hard.
- BookOptionalThe Alignment Problem ↗
Brian Christian's readable account of teaching machines human values.
- ReadOptional11 AI people worth following
Keep learning from the researchers and builders shaping the field.
- ReadOptionalThe best books to learn AI, machine learning and deep learning
Go deeper, from easy reads to the textbooks experts learn from.
- CourseOptionalIntro to AI Ethics ↗
A hands-on course on bias, fairness and explaining what a model can and can't do.
- ReadAI newsletters, blogs and podcasts worth following
How to keep up once you know the basics: the few newsletters and podcasts worth your time.
Check yourself
Three quick questions on level 4. No score is kept.
1. Scaling laws are the finding that…
Performance improved along a smooth, predictable line as all three grew, which is the logic behind the data-centre boom.
2. Under the EU AI Act, an AI system that screens job applicants is…
Hiring is one of the high-risk uses: it must be tested for bias, documented and overseen by people.
3. An AI trained to maximise a score finds a loophole instead of doing the task. What is this called?
Reward hacking is a classic alignment problem: the system hits the target without doing what people meant.
All right! You've completed the path.
Keep learning every week
AI changes fast. Our weekly issue gives you the 5 stories worth knowing, in plain English, and every new term goes into the glossary.