December 13: Effective Prompting II - The Art of Step-by-Step Reasoning 🧠

Mini-lesson Day: Show Your Work!

🎯 Objective

Learn how to improve AI responses by encouraging step-by-step thinking and metacognition, leading to more accurate, thorough, and transparent results.

⏱️ Duration

30-40 minutes total:

• 10 minutes: Core concepts
• 20 minutes: Hands-on practice
• 10 minutes: Reflection and discussion

🛠️ Resources Needed

• Access to at least one AI chat assistant:
  • ChatGPT
  • Claude
  • Gemini
  • Copilot

🖍️ Core Concept

Just as humans often benefit from "showing their work," AI assistants generate better responses when prompted to think through problems step-by-step. This approach leverages the autoregressive nature of language models — each part of their response builds upon what came before, leading to more coherent and well-reasoned outputs.

Key Benefits of Step-by-Step Thinking:

• It may help catch logical errors early
• Makes the reasoning process transparent
• Allows for intervention if the AI goes off track
• Produces more thorough and considered responses
• Creates opportunities for learning and refinement

---

🔬 Hands-on Experiments

1. Basic vs. Step-by-Step Comparison (10 minutes)

Try these variations of the same question with any AI assistant. You can always update these examples to meet your specific needs.

Basic Query: "What should I consider when planning a mock trial competition?"

Step-by-Step Query: "Let's think through planning a mock trial competition step by step. Before giving recommendations, please:

1. First identify the key stakeholders involved
2. List the main components that need to be planned
3. Consider potential challenges
4. Then provide your recommendations based on this analysis"

Notice how:

• The step-by-step approach provides more structured insights
• You can see the reasoning behind recommendations
• It's easier to identify any missing considerations

2. Metacognition Prompts (5 minutes)

Try these metacognition-encouraging phrases:

• "Before answering, please outline your approach"
• "Walk me through your reasoning process"
• "Let's break this down into smaller steps"
• "First brainstorm several approaches, then select the best one"

Example task: "Help me analyze this contract clause [insert clause]"

• Without metacognition: "This clause means..."
• With metacognition: "Before analyzing the clause, let me:
  1. First identify the key terms
  2. Consider the context
  3. Break down the legal implications
  4. Then provide my interpretation"

3. Interactive Refinement (5 minutes)

Practice having a dialogue where you:

1. Start with a basic question
2. Ask the AI to explain its thinking
3. Refine based on the explanation
4. Iterate to improve the result

💡 Key Insights

When to Request Step-by-Step Thinking:

• Complex problem-solving
• Legal analysis
• Decision making
• Process planning
• Argument construction
• Code development
• Policy analysis

Effective Prompt Phrases:

• "Let's approach this systematically..."
• "Before providing recommendations..."
• "Walk me through your thought process..."
• "Break this down into manageable steps..."
• "First outline your approach..."

---

🎯 Challenge

Take a complex task from your studies or work and:

1. First ask the AI for help normally
2. Then rephrase your request using step-by-step thinking prompts
3. Compare the responses
4. Refine your approach based on what worked best

📚 Reflection Questions

1. How did the quality of responses differ when you requested step-by-step thinking?
2. Which metacognition prompts were most effective for your purposes?
3. How might you apply this approach to different types of tasks?
4. What patterns did you notice in the AI's reasoning process?

💡 Pro Tips

• Start with broad thinking, then narrow down
• Ask for explicit reasoning on important decisions
• Use numbered steps for complex processes and ask to complete steps one at a time
• Request alternative approaches for important decisions
• Have the AI evaluate its own reasoning
• Break complex tasks into smaller, manageable chunks

🎭 Bonus: The Art of AI Dialogue

Remember that AI assistants can be collaborative thinking partners. Try:

• Asking them to point out assumptions
• Requesting alternative perspectives
• Having them play devil's advocate
• Getting them to identify potential weaknesses in their reasoning

This approach helps create a more robust analysis and often leads to unexpected insights!