December 10: Foundations of Clear Prompting 🔮

Mini-lesson day!

🎯 Objective

Master the fundamental principles of writing clear, effective prompts for AI assistants - essential skills for academic, professional, and personal success.

🛠️ Resources Needed

🖍️ Core Concepts

The Golden Rule of Prompting

As noted in Anthropic's developer documentation:

Show your prompt to a colleague with minimal context on the task.
If they're confused, the AI will likely be too.

This approach, recommended by Anthropic's AI research team, helps identify:

  • Unclear instructions
  • Missing context
  • Unstated assumptions
  • Ambiguous terminology

Basic Prompt Framework

Simple questions like "What's the capital of France?" or "Define photosynthesis" work perfectly fine as they are. However, when you need more sophisticated or precisely tailored responses, using a structured framework can significantly improve your results. Here's a framework to use when your task requires more detail or specificity:

  1. Role: Who are you/who should the AI be?

    • "As a first-year law student..."
    • "You are an expert grant writer..."
    • "I'm a high school teacher planning..."
  2. Task: What needs to be done?

    • Clear objective
    • Specific deliverables
    • Any constraints or requirements
  3. Format: How should the response be structured?

    • Document type (essay, outline, email, etc.)
    • Length requirements
    • Special formatting needs
    • Citation style if applicable
  4. Context: What background information is relevant? What would a reasonable assistant need to know to complete the task?

    • Project purpose
    • Target audience
    • Relevant policies or guidelines
    • Time or resource constraints

🔬 Hands-on Practice

Exercise 1: Prompt Transformation

Think about a prompt you’ve tried in the past that didn’t work out like you expected. How can you improve it based on what you’ve learned so far? If you don’t have a prompt that comes to mind, try improving one of the prompts below

  • Help me plan a birthday party for my friend
  • I need to brainstorm ideas for my research paper
  • Help me generate an AI policy for my course
  • I need help organizing a professional event for students and alumni.
  • I’m preparing for a panel discussion on technology and justice. Any tips?
  • I want to plan a family weekend trip to Cape Cod. What should we do?
  • I need ideas for meal prepping for the week. Any suggestions?

If you need help or inspiration you can take a look at the examples below:

Example 1: Legal Research Vague: "Help me understand this case [insert case name]."

Clear: "I'm a 1L student at Boston College Law studying Constitutional Law. Please analyze [insert case name] attached to this conversation, focusing on:

1. The core constitutional issue

2. The Court's reasoning

3. Its significance for judicial review

[Important: Attach the case as a PDF or copy and paste relevant sections as part of the conversation]

"

Example 2: Professional Writing Vague: "Write a memo about the new work from home policy."

Better: "As HR Director at [insert organization here], I need to communicate our updated hybrid work policy. Here are our current policy documents for reference:

Please draft a memo that:

  1. Explains the 3-day in-office requirement
  2. Outlines flexibility options, referencing our department guidelines
  3. Addresses common concerns identified in our recent survey
  4. Includes next steps, directing staff to our HR procedures portal Use a professional but warm tone, max 500 words, with clear section headings."

Exercise 2: Cross-Platform Testing (15 minutes)

Take one of your improved prompts and test it across different AI platforms. Note how responses vary in:

  • Level of detail
  • Tone and style
  • Follow-through on instructions
  • Additional questions asked
  • Assumptions made

💡 Key Insights

Common Pitfalls to Avoid:

  • Assuming shared context
  • Providing vague or subjective criteria
  • Omitting format preferences (especially when you have them in mind!)
  • Forgetting to specify audience

Elements of Strong Prompts:

  • Clear objectives
  • Specific requirements
  • Measurable outcomes
  • Relevant context
  • Format guidelines

🎯 Bonus Challenge

  1. Choose a real task from your work, studies, or personal life
  2. Write an initial prompt
  3. Have an AI assistant help you refine it based on the content from this lesson (just copy and paste this material as part of your prompt)
  4. Ask for the process to be iterative
  5. Iterate on the prompt until you’re satisfied and try it out in a new chat

📚 Reflection Questions

  1. What assumptions did you discover in your prompts?
  2. How did collaboration with an LLM change your approach?
  3. Which elements of the framework were most helpful?
  4. What differences did you notice between AI platforms?

💡 Pro Tips

Using References in Prompts

LLMs can accurately use and reference links in your prompts! When including links:

  • Simply reference the link by copying and pasting, or, use markdown formatting: [link text](https://url.com)
  • Provide context about why each link is relevant
  • If the model uses your link in its output, always verify that the AI has accurately linked to source material

General Tips

  • Keep a prompt journal of effective examples
  • Test complex prompts with smaller tasks first
  • Use numbered lists for multiple requirements
  • Be explicit about any constraints or preferences
  • Consider creating templates for common tasks

🔄 Looking Ahead

Start collecting examples of AI responses you particularly like - they'll be valuable later when we explore how to create custom AI personalities and behaviors! Later in the week, you'll get access to a special AI assistant designed to help you build your own custom AI assistants. Think about interactions or responses you've received that you'd love to replicate or build upon.

A watercolor painting of a snowman on ice skates

Made with Midjourney: A small, simple watercolor snowman with a cute smil, the snowman is wearing ice skates, soft pastel colors, black background. --chaos 20 --ar 1:1 --style raw --sref https://s.mj.run/C04IY4x3NTY --personalize us3j9yo --v 6.1