Meeting 2: Prompt Engineering and Integration Brainstorming

Announcements

None today. Hurray!

Discussion Outline

  1. Pedagogical Use Cases of AI
  2. Prompt Engineering Strategies for Learning
  3. Exploring Generative AI Providers for Canvas Integration
  4. Group Reflection & Discussion

Case Study: Not All Chatbots Teach

  • Recent study: Not All Chatbots Teach
  • Compared a structured, outcome-aligned chatbot vs. a generic LLM wrapper
  • Key finding: Structured AI tools significantly improved student performance and clarity in software testing tasks

Implications for Teaching

  • Structure and scaffolding matter
  • Alignment with learning outcomes is crucial
  • Promotes metacognition and reduces surface-level AI use

Reflect & Discuss

Questions:

  • Where have you seen AI help or hinder student learning?
  • What risks or opportunities do you see in using AI inside your own course context?

Prompt Engineering for Learning

  • Prompts = Pedagogical design space
  • AI output quality hinges on the prompts we design

Using Prompts as Guardrails

  • Instructor-written prompts can:
    • Set boundaries (academic integrity, tone, focus)
    • Scaffold learning (step-by-step guidance, Socratic questioning)
    • Encourage metacognition (reflect on process)

Prompt Examples

  • “Ask the student questions to deepen their understanding of X.”
  • “You are a peer reviewer. Give friendly but rigorous feedback on their code.”
  • “Use the Feynman Technique to help the student clarify their concept.”

Reflect & Discuss

Questions:

  • What kind of prompts might support your students’ learning?
  • How can prompt templates become reusable teaching assets?

Exploring Generative AI Providers

  • Let’s compare the top options for integration and customization:

Major Providers

Provider Notes
OpenAI GPT-4o, API access, supports embeddings, used widely
Gemini Google’s LLM, integrated into Google Workspace, strong data privacy controls
Anthropic Claude Emphasizes safety and guardrails
Cohere, Mistral, HuggingFace Research/community focused, fine-tuning support

Building Custom Chatbots for Canvas

🛠️ Options include:

  • OpenAI Functions + Flask/FastAPI backend
  • Google Cloud Vertex AI with Gemini + Dialogflow
  • Custom wrappers using LangChain or Semantic Kernel
  • Embed into Canvas via LTI tools, iframes, or static tools

Reflect & Discuss

Questions:

  • What role could a chatbot play in your course?
  • Would it be formative, summative, exploratory?
  • What barriers do you foresee?

Wrap-up

Today’s Themes:

  • Pedagogical framing is essential for AI tools
  • Prompting is teaching
  • Integration is possible — and powerful — in Canvas

Next Steps

  • Share a prompt or idea in our collaborative doc
  • Try one AI tool this week — reflect on use
  • Prepare for Next Week: Deeper Prompt Engineering and use case design