AI Road Trip.

One road from “what is AI?” to production systems

AI Road Trip compresses the shape of an entire AI education into eighteen questions. The road is not decoration: it runs through three territories that mirror the three tracks of the course, and the questions get harder as you drive. Where you break down tells you exactly where your understanding thins out.

The three territories

• Stops 1–6 · Beginner country. What AI actually is, how machine learning differs from traditional programming, what narrow AI means, what a large language model is, why hallucinations happen, and what prompt engineering really involves. No maths, no code.
• Stops 7–12 · Machine learning hills. Gradient descent, overfitting, convolutional neural networks, K-means clustering, backpropagation, the F1 score, random forests, and word embeddings — the working vocabulary of an applied practitioner.
• Stops 13–18 · Deep AI mountains. Transformer architecture, query-key-value attention, RLHF, LoRA fine-tuning, mixture-of-experts routing, retrieval-augmented generation, KV caching, and prompt injection — what people building AI systems argue about.

Why the difficulty climbs

AI concepts are strongly hierarchical. You cannot meaningfully understand retrieval-augmented generation without understanding embeddings; embeddings make little sense without a grasp of how models represent meaning numerically; and that depends on knowing what training does in the first place. Presenting questions in dependency order means a wrong answer is diagnostic rather than random — it marks the exact rung of the ladder where the foundation gives way.

Most people find they drive comfortably through beginner country, slow noticeably in the hills, and stall somewhere in the mountains. That is the expected shape, and it is useful information: it tells you which track to start with instead of guessing.

Potholes are the point

A wrong answer costs a heart, keeps your car where it is, and swaps in a different question from the same territory. You cannot brute-force the same question until it yields, and you cannot coast past a concept you do not know. Three potholes and the engine gives out.

The design borrows from retrieval practice, one of the most robustly supported findings in learning science: attempting to recall something — and especially failing to, then seeing the right answer — builds far more durable memory than re-reading an explanation. The eighteen stops are drawn from a pool of twenty-four questions, so repeat drives are not memorisation drills.

Reaching the finish line

Complete all eighteen stops and you have demonstrated working familiarity with the full arc of modern AI, from definitions to production concerns. That is the point at which the certificate becomes meaningful: not as a credential, but as a marker that you covered the territory. If you break down early, the flashcards and glossary are the fastest routes back.

Start the underlying course at Module 01: What Even Is AI?, or browse all 34 modules from the homepage.