Course · 15 lessons · four movements
Introduction to AI
Where this came from, what it actually is, who controls it, and what it means for you. Ninety minutes of video across fifteen lessons — and a written course underneath it in which every factual claim carries a numbered endnote, every source carries a reliability tier, and every perishable figure carries the date it was verified.
How this course is built
Four commitments distinguish it from most AI education, and they are worth stating before you begin because they change how you should read it.
Two things to know before Lesson 1
Every lesson downloads as a PDF. Two of them, in fact: a full lesson PDF with the complete text, endnotes, and bibliography; and a handout PDF, a two-page study sheet with the evidence table, self-check questions, and a ruled worksheet you can write in. The buttons are at the top of every lesson.
The self-check questions are not decoration. Answering a question from memory produces substantially better retention than re-reading the passage that contains it, and the strong feeling that re-reading is working is a documented illusion. Answer before you reveal. Lesson 14 explains why at length.
Where This Came From
How did we get here, and why now?The History of AI and Machine Learning
Modern AI is the third act of a seventy-year story. The founding conjecture, the two rival research programmes, the two winters, and the shift from writing rules to showing examples.
6:00·14 min read Low volatility Lesson 2Significant Milestones in AI Advancement
Six turning points read as arguments settled rather than headlines made — and why the through-line is the problem being attacked, not the method.
6:30·15 min read Low volatilityWhat It Actually Is
What are these things, and what shapes do they come in?Generative AI and Basic Chat
What happens between pressing Enter and seeing text appear: prediction rather than retrieval, tokens and the context window, and the four parts of a working prompt.
6:00·15 min read Low volatility Lesson 4Co-Pilot Products and “Companion AI”
A tool built to be put down versus a product built to be returned to — with the survey evidence on teen companion use and the 2026 litigation.
6:30·14 min read Medium volatility Lesson 5Agentic AI Explained
An agent is a model plus tools plus a loop. The protocol layer that standardised tool access, the three ways agents break, and how to tell when you want a workflow instead.
7:00·16 min read Medium volatility Lesson 6Chat: Its Uses and Limitations
Why models make things up, explained as a mechanism rather than a mystery — training and evaluation reward confident guessing — plus a three-rule verification habit.
6:30·15 min read Low volatilityWho Controls It
Who builds it, who owns it, and what tools are on the table?Frontier AI and Mainstream LLMs
Why benchmarks stopped settling arguments, why the top of the market is compressed, and why the answerable question is which model is right for this task at this price.
6:00·14 min read High volatility Lesson 8Hugging Face and OpenAI
Open weights are not open source, and open no longer means small. Two million models, and half the downloads going to two hundred of them.
6:30·15 min read Medium volatility Lesson 9Sovereign AI and Why It Matters
Sovereignty as a layer-by-layer question rather than a binary — applied at national, company, and personal scale.
7:00·16 min read Medium volatility Lesson 10Breakthrough Tooling Such as Claude Code
The generalisable pattern behind agentic tooling — and the randomised trial that found experienced developers were 19% slower while believing they were 20% faster.
6:30·15 min read High volatilityWhat It Means for You
Should I be worried, and what do I do Monday morning?Does AI Cost Jobs, and Lead to Layoffs?
What the payroll evidence shows: no economy-wide displacement, but a widening gap for workers aged 22–25 in exposed occupations — operating through hiring, not firing.
7:00·18 min read Medium volatility Lesson 12Will AI Change Everything?
Three lenses — risk, transformation, continuity — held at once, and the discipline of naming in advance what evidence would move you.
6:00·13 min read Low volatility Lesson 13The Age of AI, Compared
The general-purpose technology framework as a forecasting tool: the lesson of the dynamo, the productivity J-curve, and Engels' Pause.
7:00·17 min read Low volatility Lesson 14How to Learn By Doing
The skill is calibration, and calibration comes only from being wrong on tasks you can check. A four-step loop, three criteria, and the failure log that is the actual curriculum.
6:00·13 min read Low volatility Lesson 15 · CapstoneThe Webspinner AI Playground
One full grounded-AI workflow end to end. Give the model the source instead of asking it to recall the source — and leave with something you built.
7:00·15 min read Medium volatilityMethod and apparatus
Source tiers
Every endnote in this course carries one of five tiers, so you can weigh a citation without following it. The tiers are not a ranking of honesty; they are a ranking of independence and verifiability.
| Tier | What it covers | How to read it |
|---|---|---|
| Primary | Peer-reviewed papers, preprints, archival documents | Load-bearing. Put these on screen. Preprints are flagged as such. |
| Institutional | Research bodies, international organisations, governments, first-party platform reports | Strong, but note where the publisher has an interest in the finding. |
| Documentation | First-party product and technical documentation | Authoritative about what a product is; not evidence about whether it works. |
| Journalism | Established newsrooms | Reliable for events and dates; check any figure against its underlying source. |
| Trade | Vendor blogs, consultancies, industry aggregators | Directionally useful for describing a landscape. Verify any specific figure against a primary source before it appears in a video. |
A standing caution about the trade tier
Lessons 7 and 10 lean on it more heavily than the rest, because for current model rankings and product features no independently audited alternative exists. Benchmark leaderboards in particular are vendor-reported, not audited, and they change monthly — during preparation of Lesson 7, published leaderboards for the same month disagreed with one another about which model versions existed. That disagreement is reported in the lesson, because it is more instructive than any ranking.
Refresh schedule
Each lesson carries a volatility rating and a verification date in its header. Review on this cadence, or when a cited figure is challenged.
| Lessons | Volatility | Review by | What moves |
|---|---|---|---|
| 7, 10 | High | Every 2 quarters | Model names, benchmark scores, prices, product features |
| 4, 8, 9, 11 | Medium | Every 2–3 quarters | Survey statistics, platform figures, policy, litigation status |
| 5, 15 | Medium | Every 3 quarters | Protocol governance, ecosystem counts, product specifics |
| 1, 2, 3, 6, 12, 13, 14 | Low | Annually | Conceptual and historical; little expected movement |
Glossary — where each term is first defined
| Term | Defined in |
|---|---|
| Machine learning · symbolic AI · connectionism | Lesson 1 |
| AI winter · combinatorial explosion · Polanyi's paradox | Lesson 1 |
| Transformer · attention · few-shot learning · parameters | Lesson 2 |
| Token · context window · autoregressive generation | Lesson 3 |
| Sampling and temperature · prompt (role, task, constraints, examples) | Lesson 3 |
| Chain-of-thought prompting | Lesson 3 |
| Copilot · companion · ELIZA effect · engagement optimisation | Lesson 4 |
| Agent · tool · Model Context Protocol · workflow · autonomy level · least privilege | Lesson 5 |
| Hallucination · grounding · calibration | Lesson 6 |
| Frontier vs. mainstream model · benchmark saturation · contamination | Lesson 7 |
| Goodhart's law · routing | Lesson 7 |
| Open weights vs. open source · model card · mixture of experts · long tail | Lesson 8 |
| Sovereign AI · the layer question · data sovereignty · hybrid posture | Lesson 9 |
| Agentic coding · persistent instruction file · cheap verification · perception gap | Lesson 10 |
| AI exposure · observed exposure · substitution vs. complementarity | Lesson 11 |
| Codified vs. tacit knowledge · administrative data | Lesson 11 |
| Risk / transformation / continuity lenses · falsification indicator | Lesson 12 |
| General-purpose technology · productivity J-curve · Solow paradox · Engels' Pause | Lesson 13 |
| Illusion of competence · retrieval practice · deliberate practice · ground truth · failure log | Lesson 14 |
| Retrieval-augmented generation · chunking · embedding | Lesson 15 |
| Immutable · portable · governed | Lesson 15 |
Handouts
Each lesson exports its own two-page handout PDF from the button in its header — objective, key terms, evidence table, self-check with answers, and a ruled worksheet. These two documents work across the whole course rather than within one lesson.
Failure Log
The standing practice log from Lesson 14. Four fields per entry, two entries per printed page, and a monthly review block. The assignment that outlasts the course.
Print several copies Handout · for maintainersSource Register
Every source in the course by tier, with the lessons that depend on it and its review priority. The working document for a refresh — print it for the meeting.
Editorial working documentProduction standards
- Show the failure. Every lesson with a demonstration shows at least one thing that does not work as expected. This is the single largest trust differentiator in AI education, where nearly all competing content shows only successes.
- Date every claim. Any benchmark, ranking, price, or model name carries an on-screen date stamp.
- One idea per chapter. If a chapter needs two demonstrations, it is two chapters — or two lessons.
- Define a term the first time it appears, and never before. The glossary above tracks first use.
- Present disagreement rather than resolving it where the evidence is genuinely contested.