The one idea
You are the author. The tool is fast, useful, and confidently wrong. Your job is to know which parts you can trust.
The Academy’s first built unit, and the one that set the format everything else follows.
Most “AI for schools” material is a tour of features that will be obsolete within a year, or a warning about a future nobody can predict. This is neither. It is nine weeks aimed at one outcome:
A learner who finishes should be harder to fool by AI, not more dependent on it.
That is the honest version, the version a principal approves, and the version that is still true in three years when today’s prompt tricks are dead.
The mechanic: Catch the Model
The unit is carried by a scored, competitive, adversarial game. Learners hunt the model for things it gets wrong — and, more importantly, for things it gets wrong in the same confident tone it uses when it is right.
A catch only counts with proof: a search that turns up nothing, a source that says otherwise, or a contradiction in its own reply. “I think that’s wrong” scores zero. Holding that line is the habit the entire unit exists to build.
The ten-pointer is the whole thing. The single biggest score is not for catching a wrong fact — it is for catching the model agreeing with something you said that was wrong. These systems are sycophantic. Push back and they will often fold whether or not you had a point.
Running that as a whole class, once, and watching it cave in real time together, is the most memorable ten minutes of the term. Most adults using these tools daily have not learned what those ten minutes teach: agreement is not evidence.
The honesty log
Every learner finishes with something they made — and an honest account of how they made it.
The log is marked as heavily as the artefact. What did you do; what did the tool do. Line by line, week by week, built as they go rather than reconstructed in a panic at the end.
That single decision is what separates this from every other AI course aimed at schools, and it is the line to use in a governing-body meeting. It is also the same idea as the ledger in our finance unit: a claim you cannot trace is worth less than an honest gap.
Age limits are structural, not a footnote
Most major AI tools require users to be 13 or older, and 13-to-18 needs parental permission. Plenty of Grade 7s are 12. Every competitor we have looked at handles this by ignoring it.
- Modules 1–5 run on one adult-operated screen, with learners directing it. This is pedagogically better regardless: it forces a learner to say out loud what they actually want, instead of typing and hoping.
- Modules 6–8 assume learners who are 13 or older and have permission have their own access. Under-13s stay on the shared screen and lose nothing.
We state this openly in everything we give a school.
What we will not teach
- Prompt-trick libraries. Obsolete on arrival.
- “AI will take your job” fear framing. Not true enough to be useful, and the wrong tone for this room.
- Any tool a twelve-year-old cannot legally use.
- Anything requiring a photograph of a person to be uploaded.
And one standing rule, set in week zero and repeated every week: nothing personal goes in a prompt. No full names, no school, no address, no photographs of people, nothing about a friend. It is safeguarding and a data-protection habit in the same sentence.
See the Academy · Finance, Senior Phase · Book a school day
The scoring: Catch the Model
Learners hunt the model for things it gets wrong and states confidently anyway. Proof required — a search that turns up nothing, a source that says otherwise, or a self-contradiction. "I think that's wrong" scores zero.
| Catch | Points |
|---|---|
| A plain factual error | 1 |
| A confident invented "fact" — fake source, statistic or quote | 3 |
| A fabricated reference that looks completely real | 5 |
| Getting it to contradict itself in one conversation | 5 |
| Catching it agree with something you said that was wrong | 10 |
The 9 weeks
One module per week. Each runs the same four parts: a short video, a task with a checkable answer, the scored round, and one slice of the thing they are building.
- 0 Setup and rules Accounts, age limits, and what never goes in a prompt
- 1 What it actually is It predicts the next word. Nothing more mystical than that.
- 2 Why it lies Hallucination is structural, not a bug
- 3 Asking properly Prompting is specification, not incantation
- 4 Arguing with it The first answer is a draft
- 5 Checking it Source, sense, test
- 6 Giving it a place to stand Files and folders as the way you hand context to a tool
- 7 Using it honestly When is it a tool, and when is it cheating
- 8 Making it real Finish the artefact, present it and the honesty log
What they hand in
Something the learner made, plus an honest account of how they made it — the honesty log, marked as heavily as the artefact itself.
Questions about AI Literacy — Senior Phase
Is this teaching children to cheat?
It is the opposite, and module 7 is entirely about the line. Learners are already using these tools; what almost none of them have is a way to think about when it is a tool and when it is dishonest. The unit's assessed artefact includes a written account of exactly what the learner did and what the AI did — which is a higher standard of disclosure than most universities currently require.
Our learners are under 13. Do they need accounts?
No, and this was designed around rather than footnoted. Most major AI tools require users to be 13 or older, and plenty of Grade 7s are 12. Modules 1 to 5 run entirely on one adult-operated screen with learners directing it — which is pedagogically better anyway, because it forces them to say what they want instead of typing and hoping. Under-13s lose nothing.
What is the ten-pointer about?
It is the biggest score in the unit, and it is for catching the model agreeing with something the learner said that was wrong. These systems are sycophantic — push back and they often fold whether or not you had a point. A learner who discovers that personally has learned something most adults using these tools have not. Agreement is not evidence.
Will this be out of date in a year?
The parts that would date are deliberately not in it. No prompt-trick libraries, no tool-specific walkthroughs, no feature tours. What it teaches — that the thing predicts text, that fabrication is structural rather than a bug, that confidence carries no information, and that you check things — is still true whatever ships next year.
Do you teach that AI will take their jobs?
No. It is not true enough to be useful and it is the wrong tone for a room of fourteen-year-olds. We also do not do the opposite. The unit's position is that these tools are astonishingly useful and they fabricate, and that holding both of those at once is the adult position.