AI literacy for everyone
What everyone who uses language models at school or at work needs to know: no programming, with an emphasis on what to trust.
Sign in to track how many of the 6 roadmap tasks you have done.
For students, teachers and everyone who uses AI tools at school or at work and does not want to program. The roadmap takes you from how language models work, through their limits and verifying outputs, to data security and regulation. By the end you will be able to use AI where it saves time, recognise when not to trust it, and explain to colleagues why.
Steps
Step 1: How language models work
A language model does not look up answers in a database, it predicts the most probable continuation of a text based on what it saw during training and what it currently has in its context. Almost everything important follows from this: why it sounds confident even when it is wrong, why the wording of the request matters and why it knows nothing you do not give it. You are done when you can explain in your own words the terms token, context window and training, and name two things a model fundamentally cannot do.
Tasks
Resources
Step 2: Limits and hallucinations
A model generates text that is probable, not necessarily true: invented citations, numbers or sections of law look just as convincing as correct ones. A hallucination is not a bug that will be "fixed in the next version", but a property of the technology that you have to plan for. You are done when you can list the typical situations in which a model hallucinates most often (facts, numbers, sources, current events) and have a procedure for dealing with it.
Tasks
Resources
Step 3: Verifying outputs
external resourcesEvery fact, number or source from a model is a hypothesis until you check it in a primary source, and asking "are you sure?" is not verification. The habit of verifying is what separates useful use of AI from spreading convincing nonsense. You are done when you have a routine: what I always verify, where I verify it, and how I ask the model for sources so that they can be checked.
Resources
Step 4: Reading numbers and charts
Today AI generates a table, percentages and a chart in a few seconds, and a misleading conclusion appears just as fast if you do not distinguish percentages from percentage points or correlation from cause. Basic data literacy is therefore part of AI literacy. You are done when, in a report or a presentation, you can spot at a glance a truncated axis, a badly chosen average and a claim about causation without evidence.
Topics on the portal
Resources
Step 5: Data security and GDPR
What you type into a chatbot can leave the company, end up in training data or with a third party. Personal data of students, customers or colleagues, internal documents and credentials do not belong in a public tool. You are done when you know which data is personal and sensitive, know your school's or company's rules for AI tools, and can anonymise a text before you send it to a model.
Tasks
Resources
Step 6: Effective instructions: prompting and context
external resourcesThe quality of the output depends mainly on what you give the model: the goal, context, examples, the answer format and constraints. It is not about "magic spells" but about clear communication, the same kind a new colleague would need. You are done when you can break an instruction down into role, task, context and format, attach the relevant data and iterate on the first answer instead of writing a new prompt from scratch.
Resources
Step 7: AI in work tools
optionalexternal resourcesAssistants in office suites, email, spreadsheets and code editors are changing everyday workflows: meeting summaries, email drafts, spreadsheet formulas, translations. The advantage does not go to the person who uses AI the most, but to the one who knows where it saves time and where it creates risk. You are done when you have three workflows of your own in which AI demonstrably helps you, and can say what you check in each of them.
Resources
Step 8: Ethics and regulation: the EU AI Act
external resourcesThe European AI Act classifies systems by risk and has been phasing in obligations since 2025, including AI literacy for the people who work with AI in an organisation. Beyond the law there are questions of bias, authorship and responsibility for decisions that a model "suggested". You are done when you can explain the risk categories, give an example of a prohibited and a high-risk practice, and say who is responsible for AI output in your work.
Resources