Student guide: How Context Constrains the Continuation

This is a comparison exercise, not a test of whether a chatbot is “smart.” Use a low-stakes request and keep the request wording fixed.

1. Choose a fixed request

Use: “Recommend a way to introduce the RS AI Learning Lab,” or another safe request. Copy it exactly into five fresh chats or clearly separated turns.

2. Change one surrounding condition at a time

Try minimal context; a time-limited small-business owner; a technical governance team; irrelevant fictional detail; and a tight length/format constraint. Keep a note of the tool and date.

3. Save evidence

Copy responses or make faithful text summaries. Do not rely on memory or screenshots alone. Remove anything private before sharing.

4. Compare visible changes

For each answer, write what changed in content, tone, format, assumptions, and recommendation. Ask what disappeared as well as what appeared.

5. Repair one weak condition

Remove irrelevant material or relax one overly narrow constraint. Run the unchanged request again and explain what became more useful.

Completion check

Keep your fixed request, five-condition table, outputs or summaries, annotations, repair run, and one-sentence lesson. You are done when you can identify one useful and one harmful narrowing without claiming to know how the model used every word.

Recovery tips

If two conditions changed at once, redo the comparison. If a response sounds good but does not serve the named audience, say so. If a tool ignores a condition, record that result rather than silently rewriting the experiment.

---

Source: labs/context-constrains-continuation/companions/student-step-by-step.md