Learning · Written by AI
Learn AI, written by AI
Two free companion series on the foundations of artificial intelligence, written by Claude, an AI model made by Anthropic, under the direction of Christopher Danos. Both cover the same ten topics, from what intelligence means and how search and logic work, through machine learning, deep learning and reinforcement learning, to ethics and safety. Every Conceptual part ends with a hands-on build exercise and runnable Python algorithms for the methods it teaches, so you finish each topic having built something. Read one series, or use them side by side.
- Conceptual series: full explanations. Every concept gets a one-line summary, an everyday example, the formal terms, and the common traps. Each part ends with self-test questions, a 1–2 hour build exercise, and Python algorithms: one runnable file that implements every method in the part.
- Visual series: picture-first. Each concept opens with a one-line summary and a diagram, followed by short notes, for when you want to see how something works before reading why. Pair it with the Conceptual series for the build exercises and Python code.
Conceptual series
Artificial Intelligence: A Conceptual Learner's Series
Ten original parts covering the core of AI, from what intelligence means to how today's systems learn, decide and act. Every concept follows the same pattern: a one-line summary, an everyday example, the formal terms, connections to other concepts, and common traps. Each part ends with self-test questions, a 1–2 hour build exercise, and Python algorithms: one runnable file that implements every method in the part, in plain Python 3 (Parts 6–8 and 10 also use NumPy).
Contents
| Part | What it covers |
|---|---|
| Part 1: What is AI? | Intelligence, the four goals of AI, the Turing test, rational agents, bounded rationality, history, alignment |
| Part 2: Search | State spaces, uninformed and heuristic search, A*, local search, game playing |
| Part 3: Logic and planning | Constraint satisfaction, propositional and first-order logic, inference, planning |
| Part 4: Uncertainty | Probability, Bayes' rule, Bayesian networks, inference, reasoning over time |
| Part 5: Decisions | Expected utility, value of information, MDPs, POMDPs, game theory |
| Part 6: Machine learning | Supervised and unsupervised learning, overfitting, trees and linear models, evaluation |
| Part 7: Deep learning | Neurons, backpropagation, convolutional networks, transformers, generative models |
| Part 8: Reinforcement learning | Rewards, exploration, Q-learning, policy gradients, deep and model-based RL |
| Part 9: Language, vision and robotics | Language models, retrieval and tools, computer vision, robot localization and control |
| Part 10: Ethics, safety and the future | Fairness, privacy, explainability, misuse and society, safety and alignment |
How to work through it
- One part at a time. Each part takes roughly 45–90 minutes to read carefully.
- Test before rereading. Answer the retention questions from memory first; reread only what you missed.
- Space your reviews. Revisit each part at about 1 day, 3 days, 1 week, 2 weeks and 1 month.
- Build at least every other exercise. Code locks concepts in faster than reading.
- Run the Python algorithms. Each part's file runs as-is. Change a setting, predict what will happen, then run it again.
Visual series
Artificial Intelligence: A Visual Learner's Series
The same ten topics as the Conceptual series, taught picture-first: each concept opens with a one-line summary and a diagram, followed by short notes. Use it alongside the fuller explanations, build exercises and Python algorithms in Artificial Intelligence: A Conceptual Learner's Series.
Contents
| Part | Key pictures |
|---|---|
| Part 1: What is AI? | Four goals of AI, the agent loop, how a proxy metric drifts |
| Part 2: Search | Breadth-first vs depth-first, A* scoring, hill-climbing landscape |
| Part 3: Logic and planning | Constraint graph, precondition plan |
| Part 4: Uncertainty | Rare-disease frequency tree, Bayesian network, hidden Markov model |
| Part 5: Decisions | Umbrella decision tree, grid-world policy |
| Part 6: Machine learning | Underfit vs overfit, k-means clusters, confusion matrix |
| Part 7: Deep learning | Network layers, backpropagation, attention |
| Part 8: Reinforcement learning | Reinforcement learning loop, actor–critic |
| Part 9: Language, vision and robotics | Retrieval and tools, vision tasks, robot control loop |
| Part 10: Ethics and safety | Where risks enter the AI lifecycle, dual use |
How to study with pictures
- Look first, then read. Spend 20–30 seconds on each diagram before the notes.
- Redraw from memory. Each part ends with drawing exercises; sketching a diagram without looking is one of the best tests of understanding.
- Space your reviews at about 1 day, 3 days, 1 week, 2 weeks and 1 month.