Suffolk University • Sawyer Business School

How AI Actually Works

From neurons to LLMs — demystifying the technology powering the AI revolution

Session 2 of 13
150 Minutes
40 Students
Hands-On Learning
01 // LEARNING OBJECTIVES

What You'll Learn Today

By the end of this session, you'll have both conceptual understanding and hands-on experience with the foundations of AI.

01

Understand Neural Networks

Visualize how neural networks learn patterns and make decisions through layers of mathematical operations.

02

Demystify LLMs

Understand how Large Language Models like ChatGPT are built, trained, and how they generate responses.

03

Connect to Business Value

Identify real-world business applications where neural networks and AI drive competitive advantage.

04

Build Your First AI Model

Use interactive tools to train your own classifier and experience the AI development process firsthand.

02 // SESSION AGENDA

Today's Journey

A carefully crafted progression from concepts to hands-on building.

5:00 PM - 5:25 PM

Part 1: The Big Picture

25 minutes

Opening hook with live AI demo, then dive into the brilliant 3Blue1Brown visual explanation of neural networks.

  • Live demo: What happens inside ChatGPT?
  • Video: "But What is a Neural Network?"
  • Key concept: Neurons are just "things that hold numbers"
5:25 PM - 6:05 PM

Part 2: Interactive Exploration

40 minutes

Hands-on learning with TensorFlow Playground and the stunning LLM Visualization tool.

  • TensorFlow Playground challenges (3 levels)
  • Explore: Why do deeper networks solve harder problems?
  • 3D visualization of GPT-style transformers
6:05 PM - 6:40 PM

Part 3: How LLMs Are Built

35 minutes

The complete training pipeline from internet data to helpful assistant.

  • Pre-training: Learning from the internet
  • Tokenization: Breaking text into pieces
  • Post-training: Making it useful
  • Live coding demo in Google Colab
6:40 PM - 6:50 PM

☕ Break

10 minutes

Stretch, grab a coffee, check your messages. Let the concepts settle.

6:50 PM - 7:20 PM

Part 4: Business Applications & Limitations

30 minutes

Where AI creates value — and where it fails spectacularly.

  • Customer segmentation & recommendations
  • Fraud detection & demand forecasting
  • "Jagged intelligence": Finding where LLMs fail
  • Activity: 5 prompts that break AI
7:20 PM - 7:45 PM

Part 5: Build Challenge

25 minutes

Build your own image classifier with Google's Teachable Machine.

  • Teams of 2-3 students
  • Train a 3-class classifier in 10 minutes
  • Reflect: What made it work? What didn't?
7:45 PM - 7:55 PM

Part 6: Wrap-Up & Preview

10 minutes

Key takeaways, homework assignment, and what's coming next.

  • 4 key insights to remember
  • Homework overview
  • Preview: Prompt Engineering & API calls
FEATURED VIDEO

3Blue1Brown: Neural Networks

The most intuitive visual explanation of neural networks ever created.

03 // INTERACTIVE RESOURCES

Tools & Learning Materials

Everything you need to explore, experiment, and build.

Interactive

TensorFlow Playground

Play with a real neural network in your browser. Adjust neurons, layers, and watch learning happen in real-time.

Explore now
Interactive

LLM Visualization

Stunning 3D walkthrough of GPT architecture. See every layer, attention head, and matrix operation.

Explore now
Build Tool

Teachable Machine

Train your own image, sound, or pose classifier with zero code. Export to use in your projects.

Start building
Video

Karpathy: Deep Dive into LLMs

3.5 hour masterclass from the former Tesla AI Director. The definitive explainer for how ChatGPT works.

Watch now
Tool

OpenAI Tokenizer

See exactly how LLMs break text into tokens. Essential for understanding prompts and context limits.

Try it
Code

Google Colab

Free cloud Python notebooks with GPU access. Write and run code without installing anything.

Open Colab
04 // HANDS-ON ACTIVITIES

Learning by Doing

Three interactive challenges to build your intuition.

Activity 1

TensorFlow Playground Challenge

Navigate to playground.tensorflow.org and complete these three challenges:

1

Warm-up: Circle Dataset

Classify the circular data with just 1 hidden layer. How many neurons do you need?

2

Challenge: XOR Problem

Why does 1 layer fail on XOR? Add layers until it works. What's the minimum architecture?

3

Competition: Spiral Dataset

Race to get the lowest test loss! Screenshot your best result and architecture.

Activity 2

Breaking the AI: Jagged Intelligence

Test these prompts on ChatGPT, Claude, or Gemini. Document where they fail:

Test Prompts
1. "How many r's are in the word 'strawberry'?"

2. "What is 47 × 38?" (Don't let it use tools)

3. "Spell 'encyclopedia' backwards"

4. "Count the number of words in: 'The quick brown fox jumps over the lazy dog'"

5. "Who won the Super Bowl in February 2026?"

Discussion: Why do these fail? What does this tell us about how LLMs "think"?

Activity 3

Build Your Own Classifier

Use Teachable Machine to train a custom image classifier in 10 minutes:

1

Choose 3 Objects

Phone, notebook, pen — or anything around you. You need distinct categories.

2

Collect Training Data

Record ~30 samples per class using your webcam. Vary angles and lighting.

3

Train & Test

Click "Train Model" and test with new images. What's your accuracy?

4

Reflect

What made classification easy or hard? How would this apply to a business problem?

05 // KEY TAKEAWAYS

What to Remember

Neural networks are mathematical functions that learn patterns from data through layers of simple operations.

LLMs predict the next token — they don't "know" things, they simulate plausible text based on patterns.

Business value = pattern recognition — recommendations, fraud detection, forecasting all use these same principles.

AI augments, not replaces — human oversight remains essential for verification and judgment.

06 // HOMEWORK

Before Next Session

Complete these tasks before our next class to prepare for prompt engineering.

Watch: Karpathy's "Deep Dive into LLMs" — at minimum, watch 0:00-45:00 (pretraining) and 1:30:00-2:00:00 (hallucinations & psychology)
Practice: Get your best score on TensorFlow Playground's spiral dataset. Screenshot your architecture and loss value.
Think: Identify one business problem in your area of interest that could benefit from pattern recognition AI. Be ready to share.
Coming Next

Session 3: Prompt Engineering & API Calls

Learn the art and science of getting AI to do exactly what you want. We'll write our first API calls and build a simple AI-powered application.

Prompt DesignAPI IntegrationBuild Your First App