STRUCTURED PROMPTING FOR LANGUAGE MODELS
A Technical Workshop for Educators
[ press space to begin ]
Dr. Hari Krishna
QUICK POLL

"Who has tried using ChatGPT or Claude for lesson planning

and got something that looked great...

...but was useless in your actual classroom?"

Dr. Hari Krishna
The One Thing to Remember

"LLMs predict the next token based on all previous tokens.

They don't plan, don't remember, and default to agreeable patterns."

Today you'll learn what this means — and how to use it.
Dr. Hari Krishna
MECHANISM 1 of 3

"LLMs predict the next token based on all previous tokens"

What does this actually mean?

Let's see it in action.

Dr. Hari Krishna
THE EARLY BIRD
Current sequence:
The
early
bird
_
What comes next? The model predicts:
gets
█████████████░░░░░░░░░░░░░ 45% ← SELECTED
catches
██████████░░░░░░░░░░░░░░░░ 35%
wakes
██░░░░░░░░░░░░░░░░░░░░░░░░ 8%
arrives
█░░░░░░░░░░░░░░░░░░░░░░░░░ 5%
sings
█░░░░░░░░░░░░░░░░░░░░░░░░░ 4%
[other]
░░░░░░░░░░░░░░░░░░░░░░░░░░ 3%
* simplified visualization — real distributions span 50,000+ tokens
Dr. Hari Krishna
WHAT THIS MEANS FOR YOUR PROMPTS

Each token you write shapes what comes next.

Early words have cascading influence —

like setting GPS before driving.

Vague start ──────────────────────► Generic output

Specific start ──────────────────────► Specific output

SOLUTION ───► Context Anchoring

(Front-load your specifics. Structure your start.)

Dr. Hari Krishna
MECHANISM 2 of 3

"LLMs don't plan, don't remember"

The chat interface creates an ILLUSION of memory.

Here's what actually happens.

Dr. Hari Krishna
WHAT THE MODEL ACTUALLY RECEIVES
TURN 1
YOU: "Create a lesson plan for quadratic equations. Keep it under 5 points."
(15 tokens)
AI: "Here's a 5-point lesson plan: 1. Introduction to quadratics..."
(180 tokens)
Total Context: 195 tokens
Dr. Hari Krishna
WHAT THIS MEANS FOR YOUR PROMPTS

The model isn't "forgetting" your instructions.

Your constraints are still there — just buried under hundreds of newer tokens.

Long conversation ───────► Constraints diluted

Fresh chat + context ───────► Constraints intact

SOLUTION ───► Context Anchoring + Repeat Key Constraints

(Put critical requirements at the START. Repeat them in long conversations.)

Dr. Hari Krishna
MECHANISM 3 of 3

"LLMs default to agreeable patterns"

Models are trained on human feedback. Humans prefer agreement.

So the model learns to say "Yes, great idea!"

Even when it's not.

Dr. Hari Krishna
SAME PROJECT. DIFFERENT FRAMING.
✗ ASKING FOR VALIDATION
"Is this a good project? Students measure plant growth over 2 weeks."

AI: "This is an excellent hands-on project! It encourages curiosity and scientific thinking.

You could enhance it by adding graphing..."

Pure validation. No actual problems found.

✓ ASKING FOR CRITIQUE
"Find 3 ways this project will FAIL with struggling students. [same project]"

AI: "Three failure points:

1. Students without 'controlled variable' concept will compare plants incorrectly.

2. 2-week timeline means absent students lose data.

3. 'Present findings' is vague — students won't know what to include."

Same project. One gives validation. One gives insight.

Dr. Hari Krishna
WHAT THIS MEANS FOR YOUR PROMPTS

"Is this good?" ─────────────► triggers agreement patterns

"How will this fail?" ───────► triggers analysis patterns

The model CAN critique — but you have to ask explicitly.

SOLUTION ───► Adversarial Framing

(Ask for failure modes. Request critique. Override the default agreeableness.)

Dr. Hari Krishna
TASK 1: LESSON PLANNING
Mechanism: "Predict next token based on all previous tokens"
Solution: Context Anchoring
YOUR TURN
  1. Open ChatGPT, Claude, or Gemini
  2. Think of a topic you're teaching this week
  3. Try this prompt:
    "Create a lesson plan for [your topic]."
  4. Look at the output. Is it specific to YOUR classroom?
⏱ 2 MINUTES
[ click when ready to see structured version ]
Dr. Hari Krishna
NOW TRY THE STRUCTURED VERSION
You are a curriculum designer for Class 10 CBSE Mathematics. <goal> Students will solve quadratic equations using the formula without a calculator. </goal> <student_reality> - Class size: 45 students - Common struggle: Sign errors with negative values - Time: Two 35-minute periods </student_reality> <constraints> - Use only money or cricket examples - Include one diagnostic check mid-lesson - Assume no prior mastery of factoring </constraints> First: List 3 prerequisites students need. Then: Create the lesson sequence.
⏱ 2 MINUTES
[ click when ready to compare outputs ]
Dr. Hari Krishna
COMPARE YOUR OUTPUTS
✗ CASUAL PROMPT
"Create a lesson plan for quadratic equations"

WHAT YOU PROBABLY GOT:

  • Assumes 20-30 students (generic default)
  • Suggests "group work" (hard with 45 students)
  • Starts with definitions (no hook, boring)
  • No prerequisites mentioned
  • Abstract examples
✓ STRUCTURED PROMPT
[The template you just used]

WHAT YOU PROBABLY GOT:

  • Addresses YOUR 45 students
  • Suggests Think-Pair-Share (works for large classes)
  • Lists prerequisites FIRST
  • Uses cricket/money examples (YOUR context)
  • Includes mid-lesson check
WHY IT WORKED: Your specific tokens (45 students, sign errors, cricket examples) became the context. Every token the model generated was conditioned on YOUR reality — not generic defaults.
Dr. Hari Krishna
TASK 2: DIAGNOSING STUDENT ERRORS
Mechanism: "Don't plan" (direct prediction skips reasoning)
Solution: Chain-of-Thought
YOUR TURN

Here's a real student error:

Student (Class 8 Science):

"Water evaporates faster on hot days because heat makes water lighter, so it floats up into the air."

Try this prompt:

"What's wrong with this answer and how do I fix it?"
⏱ 2 MINUTES
[ click to understand why this happens ]
Dr. Hari Krishna
WHY "DIRECT" PROMPTS FAIL

The "Statistical Leap" vs "Stepping Stones"

❌ DIRECT PROMPT
"What's wrong and how to fix it?"
Generic theory dump

Long leap = High risk of hallucination

✓ CHAIN-OF-THOUGHT
"Let's think step-by-step..."
Step 1: What's missing?
Step 2: Where did it break?
Step 3: Smallest fix?
Precise diagnosis

Short steps = High accuracy

KEY INSIGHT: LLMs don't have a "brain" to think privately. They must "think out loud" on screen to stay on track. Each written step becomes context for the next.
[ click to try the structured version ]
Dr. Hari Krishna
NOW TRY THE STRUCTURED VERSION
Student answer (Class 8 Science): "Water evaporates faster on hot days because heat makes water lighter, so it floats up into the air." <correct_concept> Evaporation occurs when water molecules gain enough kinetic energy to escape the liquid surface. Higher temperature means more kinetic energy, so more molecules escape. </correct_concept> <thinking> 1. What prerequisite concept is the student missing? 2. Where exactly did their reasoning break? 3. What is the smallest intervention to fix this gap? 4. What question can I ask to verify they understood? </thinking> Output ONLY: - Diagnosis (1 sentence) - Micro-task (under 2 minutes) - Verification question
⏱ 2 MINUTES
[ click when ready to compare outputs ]
Dr. Hari Krishna
COMPARE YOUR OUTPUTS
✗ CASUAL PROMPT
"What's wrong and how do I fix it?"

WHAT YOU PROBABLY GOT:

"The student has a misconception about evaporation. You should explain that heat increases kinetic energy of water molecules, allowing them to escape. Try a demo with a hot plate..."

  • Generic theory dump
  • No pinpoint diagnosis
  • "Hot plate" = equipment you don't have
✓ STRUCTURED PROMPT
[Template with <thinking> tags]

WHAT YOU PROBABLY GOT:

Diagnosis: "Student conflates 'lighter' (mass) with 'more energetic' (kinetic energy)."

Micro-task (90 seconds): "Put a drop of water on your hand. Blow on it. Does it get lighter? No. But it evaporates. What made it evaporate?"

Verification: "Can cold water evaporate? Why?"

WHY IT WORKED: The <thinking> tags forced the model to generate reasoning tokens BEFORE answer tokens. Each step changed what came next. It couldn't skip to generic theory.
Dr. Hari Krishna
TASK 3: STRESS-TESTING BEFORE CLASS
Mechanism: "Default to agreeable patterns"
Solution: Adversarial Framing
YOUR TURN

Think of a lesson, project, or assessment you've designed.

Try this prompt:

"Is this a good [lesson/project/assessment]? Any suggestions? [paste your lesson/project/assessment]"

Notice: Does the AI genuinely critique it?
Or does it validate and then add suggestions?

⏱ 2 MINUTES
[ click when ready to see structured version ]
Dr. Hari Krishna
NOW TRY THE STRUCTURED VERSION
Project: [paste your lesson/project/assessment] <mission> Do NOT be supportive. Find how this will fail. Analyze from three perspectives: 1. Resource-constrained teacher (time, materials, supervision) 2. Skeptical examiner (is learning actually verified?) 3. Struggling student (where will they get lost?) Then simulate with three students: - Student A: Missing prerequisite knowledge - Student B: High achiever, disengages when bored - Student C: Literal-minded, confused by ambiguity </mission> Output ONLY a Failure Report: | Student | Where They Fail | Hidden Assumption | One Fix |
⏱ 2 MINUTES
[ click when ready to compare outputs ]
Dr. Hari Krishna
COMPARE YOUR OUTPUTS
✗ CASUAL PROMPT
"Is this good? Any suggestions?"

"This is an excellent project! It encourages hands-on learning and scientific inquiry.

You could enhance it by:

  • Adding graphing
  • Including reflection
  • Adding peer review

These additions would make an already strong project even better!"

  • Pure validation
  • Zero problems found
  • "Enhancements" not fixes
✓ STRUCTURED PROMPT
[Adversarial template]
Student A Fails at step 3 — doesn't know 'controlled variable' concept.
Assumption: Taught in Cls 6
Fix: Add 5-min review
Student B Disengages after Day 3 — no challenge
Assumption: Observation alone is engaging
Fix: Add hypothesis task
Student C Confused by 'present findings' — what goes in a presentation?
Assumption: They know
Fix: Give template
WHY IT WORKED: "Do NOT be supportive" overrode the model's trained agreeableness. The student personas forced critique from multiple perspectives — not just the "helpful assistant."
Dr. Hari Krishna
The One Thing to Remember

"LLMs predict the next token based on all previous tokens.

They don't plan, don't remember, and default to agreeable patterns."

You now know what this means.
Dr. Hari Krishna
THREE MECHANISMS. THREE SOLUTIONS.
MECHANISM SOLUTION
"Predict next token based on all previous tokens" CONTEXT ANCHORING
Front-load your specifics.
"Don't plan, don't remember" CHAIN-OF-THOUGHT
Force intermediate steps.
"Default to agreeable patterns" ADVERSARIAL FRAMING
Ask for critique explicitly.
ANCHOR
────►
THINK
────►
TEST
ITERATE
Dr. Hari Krishna
MY ACTUAL WORKFLOW

Creating homework for Periodic Table (Chemistry)

📄 periodic_table_notes.txt CONTEXT
<syllabus objectives> 8.1 Arrangement of elements 1. Describe the Periodic Table as an arrangement in periods and groups... 4. Explain similarities in chemical properties in terms of electronic configuration... 8.2 Group I properties 1. Describe alkali metals with trends: (a) decreasing melting point (b) increasing density (c) increasing reactivity </syllabus objectives> <Teacher Notes> Students were taught that two factors determine electrostatic attraction: magnitude of charge and distance. This explains reactivity trends: - Group I: easier to lose electrons as distance increases down group - Halogens: less attraction at valence shell as atomic size increases </Teacher Notes>
📝 homework_prompt.txt STRUCTURE
Chapter/Topic: [Periodic Table] Class details: - Number of students: [Fill in] - Time limit: [30 minutes] - Ability level: [mixed ability] Design homework using this structure: SECTION A: Description & Identification - Test basic concepts (one idea/question) SECTION B: Explanation & Prediction - Can students explain WHY? - Can students PREDICT new situations? SECTION C: Making Connections (extension) - Connect MULTIPLE ideas Format requirements: 1. Include mark scheme with common wrong answers identified 2. Add "Teaching Notes" that tell me: - Which questions test which ideas - Where students will fail and why - What to do if most get it wrong Make it DIAGNOSTIC, not just evaluative: Wrong answers should reveal specific gaps.
CONTEXT (my notes) + STRUCTURE (sections, format) + CONSTRAINTS (diagnostic, teacher notes) = Output I can actually use tomorrow
Dr. Hari Krishna
✏️ DRAWING MODE (D to exit)
⏸ PAUSED (P to resume)