Learn AI, one idea at a time
From the first artificial neuron to today's agents. Short lessons with a small animation, a hands-on mini tool and a 3-question check.
- chapters
- 21
- lessons
- 222
- to finish
- ~26h
- hands-on
- 26
Your progress, track and quiz scores are saved on this device only, never sent anywhere.
Pick your path
Not sure where to start?
Choose a track. We highlight its lessons below, and the main button follows it.
Chapter 0
Start here
What AI is, in plain words, and how this course works.
Chapter 1
The dawn of AI (1943–1990)
The first ideas: artificial neurons, the Turing test, the word 'AI', and the winters.
- 1.1The first artificial neuron1943
- 1.2The Turing test1950
- 1.3The birth of 'artificial intelligence'1956
- 1.4The perceptron1958
- 1.5Symbolic AI
- 1.6Search and game playing
- 1.7ELIZA, the first chatbot1966
- 1.8The XOR problem1969
- 1.9Expert systems
- 1.10AI winters
Chapter 2
Machine learning basics
How machines learn from data: models, loss, training and testing.
- 2.1Data and features
- 2.2Labels and supervised learning
- 2.3Models and parameters
- 2.4Loss: measuring mistakes
- 2.5Gradient descent
- 2.6Learning rate
- 2.7Linear regression
- 2.8Logistic regression
- 2.9Training vs inference
- 2.10Train, validation and test sets
- 2.11Overfitting and underfitting
- 2.12Bias and variance
- 2.13Regularisation
- 2.14Accuracy, precision and recall
- 2.15The confusion matrix
- 2.16Decision trees
- 2.17Random forests and ensembles
- 2.18k-nearest neighbours
- 2.19Clustering (unsupervised)
- 2.20Dimensionality reduction
- 2.21Reinforcement learning
- 2.22Normalisation and feature scaling
- 2.23Collecting and cleaning data
- 2.24Recommendation systems
- 2.25Forecasting over time
Chapter 3 · 1986–2015
Neural networks
Neurons, weights, layers and how networks learn with backpropagation.
- 3.1A neuron: weights and bias
- 3.2What weights really are
- 3.3Activation functions
- 3.4Layers and the multi-layer network
- 3.5The forward pass
- 3.6Backpropagation1986
- 3.7The chain rule, intuitively
- 3.8Epochs, batches and steps
- 3.9Optimisers: SGD, momentum, Adam
- 3.10Vanishing and exploding gradients
- 3.11Dropout2014
- 3.12Batch and layer normalisation2015
- 3.13Why networks can learn almost anything1989
Chapter 4
The deep learning era (2006–2016)
GPUs, ImageNet, convolution, word vectors and the networks that changed everything.
- 4.1GPUs and the deep learning boom2009
- 4.2ImageNet and AlexNet2012
- 4.3Convolution
- 4.4Convolutional neural networks
- 4.5Pooling
- 4.6Residual connections (ResNet)2015
- 4.7Recurrent neural networks
- 4.8LSTM: long short-term memory1997
- 4.9Word vectors (word2vec)2013
- 4.10Autoencoders
- 4.11GANs: generator vs discriminator2014
- 4.12Sequence to sequence2014
- 4.13Attention is born2014
- 4.14Transfer learning
- 4.15Object detection2015
- 4.16Image segmentation
- 4.17AlphaGo and deep reinforcement learning2016
Chapter 5
Turning language into numbers
Tokens, vocabularies, embeddings and vectors: how text becomes maths.
- 5.1Tokens
- 5.2How tokenizers are built (BPE)2016
- 5.3Vocabulary and token IDs
- 5.4Tokens across languages
- 5.5One-hot vs dense vectors
- 5.6Embeddings
- 5.7Vectors and dimensions
- 5.8Cosine similarity
- 5.9Semantic search
- 5.10Position: how models know word order
Chapter 6 · 2017–2023
The Transformer
The 2017 architecture behind every modern language model.
- 6.1The Transformer at a glance2017
- 6.2Self-attention
- 6.3Queries, keys and values
- 6.4Softmax
- 6.5Multi-head attention
- 6.6The feed-forward block
- 6.7The residual stream
- 6.8Causal masking
- 6.9Encoders, decoders and both
- 6.10BERT: reading both ways2018
- 6.11GPT: generative pretraining2018
- 6.12Logits
- 6.13The context window
- 6.14The KV cache
- 6.15Parameters and model size
- 6.16Beyond Transformers: state-space models2023
Chapter 7
Large language models
Next-token prediction, pretraining, scaling, and how an LLM writes.
- 7.1What is a large language model?
- 7.2Next-token prediction
- 7.3Pretraining
- 7.4Scaling laws2020
- 7.5Compute and FLOPs
- 7.6Emergent abilities
- 7.7Temperature
- 7.8Top-k and top-p sampling
- 7.9Greedy decoding vs sampling
- 7.10Hallucination
- 7.11Knowledge cutoff
- 7.12Perplexity
- 7.13Base models vs chat models
- 7.14System prompts and chat roles
- 7.15Tokens, limits and cost
- 7.16Stop sequences and max tokens
- 7.17Small language models
Chapter 8 · 2021–2023
Shaping models: tuning and alignment
From a raw text predictor to a helpful, honest, harmless assistant.
- 8.1Fine-tuning
- 8.2Instruction tuning2021
- 8.3RLHF: learning from human feedback2022
- 8.4Reward models
- 8.5Constitutional AI and AI feedback2022
- 8.6Direct preference optimisation2023
- 8.7Alignment
- 8.8LoRA and efficient fine-tuning2021
- 8.9Quantisation
- 8.10Distillation
- 8.11Synthetic data
- 8.12Mixture of experts
- 8.13Open-weight vs closed models
- 8.14Benchmarks and evaluations
- 8.15AI judging AI
- 8.16Red teaming and safety testing
Chapter 9 · 2020–2022
Talking to models: prompting and context
How to ask well, and how to give a model the right information.
- 9.1Prompt basics
- 9.2Zero-shot and few-shot prompting2020
- 9.3Chain of thought2022
- 9.4Structured output
- 9.5Prompt injection
- 9.6Jailbreaks and misuse
- 9.7Context engineering
- 9.8Long context and 'lost in the middle'
- 9.9Prompt caching
Chapter 10
Using AI in daily life
Practical habits for work, study and home, for everyone.
- 10.1Prompts for everyday tasks
- 10.2Learn anything with an AI tutor
- 10.3Check AI answers
- 10.4Write and edit with AI
- 10.5AI with documents and spreadsheets
- 10.6Make images with AI
- 10.7Keep your data safe with chatbots
- 10.8Spot AI scams and fake media
- 10.9Talk to AI: voice and dictation
- 10.10When not to use AI
Chapter 11
Retrieval and memory
Giving models fresh, private and checkable knowledge.
- 11.1Retrieval-augmented generation (RAG)2020
- 11.2Chunking documents
- 11.3Vector databases
- 11.4Hybrid search and reranking
- 11.5Grounding and citations
- 11.6Memory in assistants
Chapter 12 · 2022–2024
Tools, agents and harnesses
From chatbots to AI that takes actions.
- 12.1Tool use and function calling
- 12.2The agent loop
- 12.3Reason + act (ReAct)2022
- 12.4The agent harness
- 12.5Model Context Protocol (MCP)2024
- 12.6Planning and subagents
- 12.7Multi-agent systems
- 12.8Computer-using agents2024
- 12.9Coding agents
- 12.10Permissions, sandboxes and guardrails
- 12.11Evaluating agents
Chapter 13 · 2022–2025
Reasoning models
Models that think before they answer.
- 13.1Test-time compute2024
- 13.2Thinking tokens and extended thinking2024
- 13.3Self-consistency and voting2022
- 13.4Verifiers and checking
- 13.5Reinforcement learning from verifiable rewards2025
Chapter 14 · 2020–2024
Images, sound and video
AI that sees, hears, speaks and creates media.
- 14.1Vision transformers2020
- 14.2CLIP: matching images and text2021
- 14.3Diffusion models2020
- 14.4Latent diffusion and text-to-image2022
- 14.5Speech recognition
- 14.6Text to speech
- 14.7Multimodal LLMs
- 14.8Video generation2024
- 14.9Deepfakes, watermarks and provenance
Chapter 15
Running AI in the real world
Hardware, speed, cost and AI on your own device.
- 15.1GPUs, TPUs and accelerators
- 15.2Latency vs throughput
- 15.3Time to first token and tokens per second
- 15.4On-device AI
- 15.5The cost of serving AI
Chapter 16
AI in science and the physical world
AI that folds proteins, moves robots and learns how the world works.
Chapter 17
AI and society
Fairness, privacy, safety, law and where AI is heading.
- 17.1Bias and fairness
- 17.2Privacy and training data
- 17.3Data poisoning and model theft
- 17.4Privacy-preserving AI
- 17.5Copyright and training data
- 17.6Interpretability
- 17.7Energy and the environment
- 17.8AI regulation2024
- 17.9Using AI responsibly
- 17.10AGI and the road ahead
- 17.11Recap: the AI timeline
Hands-on
set up your AI workbench
Step-by-step tutorials to run and use AI models on your own computer.
- 18.1How the hands-on tutorials workHands-on
- 18.2Terminal basics for AI workHands-on
- 18.3Set up Python and a virtual environmentHands-on
- 18.4Run an open model on your computerHands-on
- 18.5Choose a model that fits your computerHands-on
- 18.6Call your local model from codeHands-on
- 18.7The model landscape: open and closedHands-on
- 18.8Use a hosted model API safelyHands-on
- 18.9Save prompts and settings as filesHands-on
- 18.10Check your workbench end to endHands-on
Hands-on
build real projects
Build RAG over your PDFs, a local vector database, an MCP server, a small agent, and train and fine-tune models.
- 19.1Make embeddings on your computerHands-on
- 19.2Set up a local vector databaseHands-on
- 19.3Turn PDFs into searchable chunksHands-on
- 19.4Chat with your PDFs: a simple RAG appHands-on
- 19.5Test and improve your RAG appHands-on
- 19.6Build a simple MCP serverHands-on
- 19.7Connect your MCP server to an AI appHands-on
- 19.8Build a tiny agent with toolsHands-on
- 19.9Train a tiny model from scratchHands-on
- 19.10Fine-tune a small model with LoRAHands-on
Hands-on
more projects
Speech, images, data extraction, a web page for your app and more, on your own computer.
- 20.1Turn speech into text on your computerHands-on
- 20.2Make images on your computerHands-on
- 20.3Pull structured data out of documentsHands-on
- 20.4Give your RAG app a simple web pageHands-on
- 20.5Teach a model to recognise your photosHands-on
- 20.6Test your app with an AI judgeHands-on
Guides for this task
- Choose a course entry point and verify saved progress
- AI Tokens: Inspect Pieces, Spaces and Stable Teaching IDs
- Inspect candidates before continuing text
- Choose complexity using held-out error