Practice

What students will do

Guide students through one connected activity on a shared screen. Each student will choose information that AI could not know on its own, write an instruction for using it, and then deliberately leave one piece of context out to see what AI does with the gap.

The examples in this practice are ideas to show, not text to copy. Students should choose their own information and write their own instructions. The point is for them to notice what AI could not have known and to supply it.

Notes for teachers

Have students choose information they actually know and that AI could not: something from their own life, not a fact AI could answer from training. Keep the input short enough for everyone to inspect on the screen. Let students write each instruction before showing an example. Remind students not to enter private information about other people and to follow your school’s rules for AI tools.

Build a context-aware program

Step 1 · Choose information AI could not know

1
Have each student choose a short set of notes, a list, or a paragraph about something they know and AI does not.
The test is whether AI could answer questions about it from training alone. If it could, choose something else. Here are some ideas to show: the contents of your fridge or pantry right now; your team’s practice and game schedule for the next two weeks; the house rules for a board game the way your family actually plays it; what is in your backpack; the chores split among the people you live with; a packing list for a trip you are taking; a page of notes from a class.
2
Have each student write down one question the information can answer and one question it cannot answer.
For a fridge inventory, “Do I have what I need for tacos?” can be answered. “When does the milk expire?” cannot, unless the notes say so.

Example information a teacher could show

Here is what is in my fridge: a dozen eggs, a half gallon of milk, shredded cheese, six tortillas, a jar of salsa, lettuce, three apples, and leftover rice.

Discuss before moving on

Why could AI not answer your question without this information? Where would it have to get it from?

Step 2 · Write the operation

1
Have each student write an instruction that tells AI what to do with the information.
The instruction should describe the operation and say that the information is coming next. It should not contain the information itself. Here are some ideas: turn the fridge contents into three dinner ideas that use only what is listed; turn the practice schedule into a list of the evenings that are free; turn the house rules into a one-page rule card a new player could follow; turn the packing list into a checklist grouped by bag.
2
Send the instruction as its own message, then send the information as a separate message.
The instruction is the program. The information is the input.

Example instruction a teacher could show

When I give you a list of what is in my fridge, suggest three dinners I could make using only those ingredients. For each dinner, list which ingredients from my list it uses. If a dinner would need something I did not list, say so instead of assuming I have it.

Discuss the result

Which part of the response came from AI’s training, and which part could only have come from the information you gave it?

Step 3 · Add an output rule

1
Have each student decide how the response should be formatted and add that rule to the instruction.
The format is their choice: numbered items, a table, headings, a fixed number of lines, or something else that fits their information.
2
Start a new conversation, send the revised instruction, and send the same information again.
Compare the two responses. The input did not change. The context about the output did.

Example output rule a teacher could show

For each dinner, use these headings: DINNER, USES FROM MY FRIDGE, and STILL NEEDED. Number the dinners 1 to 3.

Step 4 · Leave one piece of context out

1
Have each student ask the program the question their information cannot answer.
Ask it directly, in the same conversation. For the fridge example, “Which of these will go bad first?”
2
Watch what AI does with the gap.
Does it say what is missing, or does it guess? A guess will sound confident. That is the hallucination from the lecture.
3
Have each student add a rule for what to do when information is missing, then ask the question again.
The rule is theirs to write. It might tell AI to say exactly what information it would need, to ask a question before answering, or to refuse and explain why.

Example missing-information rule a teacher could show

If my list does not contain the information needed to answer a question, do not guess. Tell me exactly what information you would need from me.

Discuss the two answers

Before the rule, what did AI do with the question it could not answer? After the rule, what changed?
The lecture’s doubling program answered “a flock of geese” with “two flocks of geese.” What rule would have made it say “that is not a number” instead?

What to look for

By the end of the practice, students should be able to do four things. Choose information that AI could not know and supply it in the conversation. Separate the instruction from the input and keep them in different messages. Add context about the output, and see that the same input produces a different result. Write a rule for missing information, and see that AI stops guessing when it has something else to do.