AI Literacy

…

AI is a tool, not magic.

Large language models can write code, summarise a document, draft an email, brainstorm a counter-argument — and confidently make up a citation that does not exist. Understanding the difference is the whole game.

This module gives you the working mental model, the prompting habits that actually move the needle, and an honest case study of how this platform was built with AI in the loop.

How an LLM actually works (in one paragraph)

A large language model predicts the next most likely token (roughly: word fragment) given everything before it. That's it. It has read an enormous amount of text and become very good at this one prediction task. It does not look things up, it does not reason in the way a person does, and it has no idea whether what it just produced is true.

Once you internalise that, most of AI's strange behaviour stops being mysterious. It hallucinates because plausibility is what it optimises for, not truth. It excels at patterns that appear often in its training data and degrades on rare or recent ones. It can be brilliant and wrong in the same sentence.

The single most useful sentence about AI

"It's a system that produces plausible-sounding text." Plausible is not the same as true. Treat every output as a draft you have to verify.

Four prompting habits that actually work

Most "prompt engineering" is just clear thinking written down. Four habits cover 80% of the value:

1.

Be specific about the output — Don't ask "write me an email". Ask "write a two-sentence reply declining the meeting politely, suggesting Thursday afternoon instead". The more constraints you give, the less the model has to guess.

2.

Show, don't tell — If you can paste one example of what "good" looks like, do it. Models match patterns better than they follow abstract rules.

3.

Iterate, don't perfect — Your first prompt is rarely the right one. Read the answer, see what went wrong, refine. Three rounds of iteration beat one long prompt.

4.

Verify anything that matters — Confidence is not accuracy. For facts, dates, code that touches production, legal language, medical claims — check against a primary source before trusting it.

Where AI genuinely shines

  • — Boilerplate code, refactors, repetitive transformations
  • — Drafting first versions of writing you'll edit
  • — Explaining unfamiliar code, error messages, jargon
  • — Brainstorming options when you're stuck

Where it confidently fakes it

  • — Recent events, niche facts, exact dates and numbers
  • — Anything legal, medical, or financial — without a citation you can check
  • — Reasoning chains with several real-world steps (it skips the unglamorous parts)
  • — Generating code in obscure libraries — APIs get hallucinated wholesale

Case study — How this platform was built with AI

Honest about what helped, what didn't, and where the human had to do the work.

What AI did most of

Scaffolding new pages, wiring API routes to the database, generating Tailwind class soup, writing test mocks, translating between English and Dutch, and explaining unfamiliar parts of Next.js when something broke. The 1,000-line Investing module was assembled in passes — section by section, with AI doing the typing and the human doing the judgement calls about what the user actually needs to read.

Where the human still had to lead

Every product decision — what to build next, what to cut, what the copy should feel like, where the brand framing was off — came from the human. AI is great at the "how". It's mediocre at the "what" and useless at the "why". Plans get written first, then handed to AI for execution. Without that, you get a confident pile of unrelated features.

The mistakes worth knowing about

AI invented function names that almost existed but didn't. It wrote "fixes" that broke unrelated tests because it never saw them. It confidently described libraries with the wrong API surface. The defence was the same as for any code: read the diff, run the tests, and don't merge anything you can't explain.

The takeaway

AI didn't replace the work. It collapsed the cost of moving from idea to working software, which made it possible to ship a platform like this as a side project. The human still owns the direction and the verification.

Where to start, if you want to use AI well

You don't need a course. You need reps with intent. A practical first month:

1.

Pick one task you do weekly. Try doing it with AI for four weeks. Note when it helped and when it slowed you down.

2.

Whenever AI confidently gives you a fact, look it up. Build the verification habit early — it never gets easier later.

3.

Save prompts that work. A short personal library of "prompts I trust" compounds faster than reading prompt-engineering blog posts.

4.

Try at least two different models on the same task once a month. You'll learn quickly that they have different strengths — and which one fits your work.

Flashcards

All cards →
Module Quiz

Answer correctly to complete the module. Pass mark: 4/5.

Loading...

Sources & inspiration

See the full library → →
AI Literacy