AI Literacy
Prompt engineering, verification, and judgement, the same discipline this site's own AI assistant and this website's own build were held to. Not a slide deck. A practical skill.
We didn't just read about this. We run on it.
Inkspire AI, the assistant on this site, was built this year with a deliberate, hard line: strict, verified facts about Inkspire itself, and genuine, substantive help on general research questions, never blurring the two. That line, and the judgement it takes to hold it, is the actual skill this programme teaches. It is a rare, genuinely true claim for an education organisation to make: we don't just teach this, we practise it in public, on this website, every day.
Four modules, in order.
Engage
What AI actually is
What a large language model actually does, explained without jargon, at the level that fits you. What "hallucination" means, and why a confident sounding answer is not the same as a correct one.
- How a model predicts the next word, and why that explains both its fluency and its failure modes
- What "hallucination" actually is: a fluent, confident answer with no underlying fact behind it, not a rare glitch
- The difference between a model that has seen a fact in training and one that is pattern matching its way to a plausible answer
- Why the same question can get a different answer twice, and what that tells you about how much to trust any single response
Create
Prompt engineering fundamentals
Specificity over vagueness, context and constraints, and treating a first answer as a draft to refine, not a final answer to hand in. Includes a research related track for literature scoping and early drafting.
- Giving a model the context it cannot infer: audience, format, length, and what "good" looks like for this specific task
- Iterative refinement: treating the first response as a draft to interrogate, not an answer to copy out
- Constraint setting: telling a model what to exclude is often more useful than describing what to include
- A research specific track: using a model to scope a literature search or structure a first outline, never to write the argument itself
Manage
Verification and judgement
Never presenting an AI generated claim as fact without checking it yourself. Understanding bias in what a model was trained on, and knowing which decisions should never be delegated to AI at all.
- A simple verification habit: every specific claim, date, citation, or statistic gets checked against a real source before it is used anywhere
- Recognising a fabricated citation: a real sounding journal name, a plausible author, a page number, none of which actually exists
- Where training data bias shows up in practice, and why a fluent answer can still quietly reflect a narrow set of sources
- A short, explicit list of decisions that should never be delegated to AI: anything medical, legal, financial, or safety related that affects a real person
Shape
Multimodal literacy
Understanding that different AI tools have different strengths, and combining them in a real workflow rather than relying on one tool for everything. The tools will keep changing; this judgement will not.
- Matching the tool to the task: a chat model for structuring an argument, a search grounded tool for anything needing a current, checkable source
- Reading across text, image, and data outputs from the same prompt, and knowing which modality actually answers the question asked
- Building a small, personal workflow: which step a model helps with, which step still needs a human, and why that boundary should not move casually
- Treating tool fluency as a moving target: the specific products will change, this module teaches the underlying judgement that outlasts any one of them
What this is not.
"Using AI is cheating."
The real distinction, consistent with our Academic Integrity position, is between using AI to understand and produce your own work faster, which this programme teaches, and using it to produce work you submit as unexamined effort, which we do not teach or permit, with or without AI in the picture.
"A good prompt is a long, complicated prompt."
Usually the opposite. A good prompt is a precise one. Module 2 is built around cutting vague words, not adding more of them.
"If it sounds confident, it is probably right."
The opposite correlation is closer to true. A model's tone carries no information about its accuracy: it is exactly as confident when it is fabricating a citation as when it is correctly stating a well known fact. Module 3 exists because this specific misconception is the one that actually causes harm.
Three ways in, depending on who you are.
Read the four modules above. Free, no registration.
Ask your mentor to add this as a module alongside your existing programme.
A workshop for your staff on responsible AI use in research writing, via our B2B pipeline.
