My GenAI Success and Challenge

Sharon Lobo

Using GenAI to Help Explain Convolutional Neural Networks (CNNs)

  • Task: Understanding how CNNs work for image classification.

  • Success: Using GenAI to explain and simplify CNN architecture in detail.

  • Prompt Example: "Please explain how a convolutional neural network works for image classification, including convolution, pooling, and fully connected layers."

Output of first prompt

Why GenAI Works Well

  1. Simplifies Complex Concepts:
    • GenAI breaks down complexity of any topic, making sure it’s focusing on the key components
  2. Provides Clear Analogies:
    • GenAI uses analogies to make difficult topics accessible and easier to understand.
    • Example: Describing convolution as “filtering an image to detect edges.”
  3. Quick Iterations:
    • You can refine explanations by asking follow-up prompts to adjust for understanding.

Breaking it down some more: Second prompt

Demonstrating GenAI’s Limitation: Generating Spatial Coordinates

Task: Generate Random Coordinates

  • Prompt: > Generate 5 random latitude and longitude points within Austin, Texas.

Prompt Output

Observed Challenge

  • Some points may fall outside of Austin.
  • GenAI cannot validate coordinates against city boundaries yet
  • Relies on “plausible values”

Why GenAI fails

  • No real-time geographic data
  • Generates random numbers without confirming validity
  • Integrate with a GIS API