Skip to main content

Advanced reasoning

These templates are useful when:

  • there are several reasonable solutions (activity design, rubrics, examples),
  • you need consistency (checklists, guidelines),
  • or it is convenient to decompose without losing restrictions (level, time, privacy, format).

Key idea: ask for alternatives + criteria + selection, not “think step by step”.

Tree of Thoughts (ToT): Explore alternatives and choose with a rubric​

You are an instructional design assistant for middle school and high school/VET.

Task:
- Design a 50-minute activity on "how to evaluate the quality of an AI response".

Context:
- Level: 10th grade
- Resources: whiteboard + optional mobile phones
- Students' final product: a verification checklist + 3 justified corrections

Instructions (ToT):
1) Propose 3 alternative designs (A, B, C) with different approaches.
2) Evaluate A/B/C using this rubric (0–2 per criterion):
   - Observable objective and evidence
   - Realistic in 50 mins
   - Quick and usable assessment
   - Inclusion (support/extension)
   - Risks and mitigations (privacy/biases)
3) Choose the best one and deliver ONLY the final version incorporating improvements.

Output (Markdown with fixed headers):
- Objective and evidence
- Materials
- Minute-by-minute sequence
- Assessment (checklist + criteria)
- Adaptations (support/extension)
- Risks and mitigations

Why it works:

  • Forces you to explore options (avoid the first mediocre idea).
  • Converts “better/worse” into observable criteria.
  • Request a final delivery ready to use.

Self-consistency: consensus for a checklist​

You are a teaching assistant.

Task:
- Create a checklist for students to evaluate an AI response before submitting it.

Constraints:
- Maximum 10 items.
- Write each item as a Yes/No question.
- No students' personal data.

Instructions (Self-consistency):
1) Generate 3 independent checklist proposals (without referring to each other).
2) Build a final checklist by consensus:
   - Keep the common items.
   - If you add an uncommon one, justify why it is critical.

Output (Markdown):
- Final checklist (≤10)
- 2 use examples (very brief)

Why it works:

  • Reduces variability and “occurrences” in a single pass.
  • Produces a short and stable list (ideal for the classroom).

Decomposition with constraints (task decomposition / prompt chaining)​

This is useful when you want a long result, but without losing global requirements.

You are an instructional design assistant for VET.

Task:
- Create a bank of 12 multiple choice questions on "security and privacy when using AI".

Context:
- Level: 1st year Basic VET
- Evaluation time: 15 mins
- Requirement: plausible distractors, clear language, no trick questions

Process (decomposition):
1) Define a blueprint with 6 subtopics (2 questions per subtopic).
2) Write the 12 questions following the blueprint.
3) Review global consistency:
   - balanced distribution by subtopic
   - appropriate level
   - no doubtful statements (if any, mark as "to verify")
   - no personal data

Output:
- Blueprint (numbered list)
- Markdown table with columns: Question | A | B | C | D | Correct | Brief justification

Why it works:

  • Control coverage (half of the agenda is not “forgotten”).
  • Maintains global restrictions until the end.