How Context Constrains the Continuation

Release mode Beta

Level
Beginner
Format
individual or pairs

Use the same request under five conditions to compare how supplied context shapes an AI continuation. You will identify details that became prominent or disappeared, distinguish content constraints from tone and format constraints, and state when narrowing became unhelpful.

Observe how what comes before a request changes what follows

What you will make

Use the same request under five conditions to compare how supplied context shapes an AI continuation. You will identify details that became prominent or disappeared, distinguish content constraints from tone and format constraints, and state when narrowing became unhelpful.

Lab materials

Use during the lab

Go deeper

Before you start

Level and format: Beginner; individual or pairs

You need: a mainstream AI chat tool, a note-taking surface, and a low-stakes task. Use non-confidential material. Different tools and new chats may produce different results; that variation is part of the observation.

The core idea

A chatbot continues from the material around a request. Relevant context can make an answer more useful; irrelevant context can distract it; excessive constraints can rule out useful options. More context is not automatically better.

Compare how context changes the continuation

Complete the activity here. Open the optional Lab Notebook beside this page when you want a place to record your work.

  1. Keep one base request stable

    Use the base request: “Recommend a way to introduce the RS AI Learning Lab.” Keep the wording fixed.

  2. Run five separate context conditions

    In separate chats or clearly separated turns, run it with: minimal context; an audience of a time-limited small-business owner; an audience of a technical team concerned with governance; irrelevant detail about a fictional event; and a tightly constrained one-slide, no-jargon, 35-word version.

  3. Record each visible continuation

    Copy each response or record a faithful summary. Do not change more than one condition at a time when possible.

  4. Compare what changed

    In a comparison table, mark what changed in content, tone, format, assumptions, and recommendation.

  5. Choose and repair one condition

    Choose the most useful answer for a named purpose. Repair one weak condition by removing irrelevant material or relaxing one constraint; run it again.

Check your result

  • Which possibilities disappeared? Which details moved to the foreground? Did a constraint change what was said, how it sounded, or both? Was a constraint ignored? Did relevance improve while exploration narrowed too far?

  • You can verify that conditions differ and compare visible outputs. You cannot prove from one small test exactly how a model internally used every word, or that a recommendation is factually sound. Check claims externally and let a human decide whether an audience description is fair.

  • A polished result is not proof that it is accurate, complete, safe, or appropriate. Keep unresolved questions visible and verify consequential claims outside the tool.

You are done when

Reflect: What was the smallest useful context? What constraint had the largest effect? What did the “best” answer leave out? Save the fixed base request, a five-condition table, selected outputs, annotations, repaired run, and one-sentence lesson. Done means you compared at least five conditions and explained one useful and one harmful narrowing. Make the table readable without color and redact chat history before sharing.

Try another version

These are optional extensions, not additional completion requirements.

Learner Map

Tracks

Modules

Help us improve this Activity

Tell us what helped, what slowed you down, or what you needed but could not find.

🎓 Give learner feedback