Module 03 · The Landscape

Key Types of AI You'll Encounter

A visual field guide to the AI systems you interact with every day — what makes each one different, and how they work under the hood.

⏱ 45 min📊 3 Diagrams🧩 4 Exercises✅ 4-Question Quiz
📖
Lesson Content
Read & Understand

Not all AI is the same. The voice assistant on your phone, the algorithm that recommends your next Netflix show, and the chatbot you use for work are three completely different systems — trained differently, on different data, for different tasks.

Large Language Models (LLMs) like Claude and GPT are trained on massive amounts of text. They predict what words should come next, developing a broad ability to write, reason, summarize, and converse. They don't look things up — they generate.

Image recognition systems use layered filters to detect edges, then shapes, then objects. Each layer builds on the last — like a visual assembly line from pixels to meaning.

Recommendation systems match you to content by finding people who liked similar things. They're powered by your behavior — clicks, watches, purchases — not your stated preferences. Predictive models look at historical data to forecast outcomes like credit scores or weather.

Key Takeaways

LLMs generate text by predicting what comes next — they don't "look things up"
Image recognition works through layers of visual pattern detection
Recommendation engines work from behavioral data, not stated preferences
Predictive models find correlations in historical data to forecast outcomes
Different AI types are NOT interchangeable — each solves a different problem
🃏
AI Type Card Set
Four major types at a glance
💬
Large Language Model
Trained on vast text — predicts next tokens to generate coherent language.
Examples: Claude, ChatGPT, Gemini
👁
Image Recognition
Layered filters extract edges → shapes → objects from pixel data.
Examples: Face ID, Google Lens, medical imaging
🎯
Recommendation Engine
Matches users to content based on behavioral patterns and similarities.
Examples: Netflix, Spotify, TikTok, Amazon
📈
Predictive Model
Uses historical data to forecast future outcomes or classify inputs.
Examples: Credit scoring, weather, fraud detection
🌐
Real-World AI Ecosystem
Products mapped to their AI type
Netflix / Spotify
Recommendation
Behavioral matching to predict what you'll enjoy
Gmail spam filter
Predictive Model
Classifies email as spam or not based on learned patterns
Face ID / Unlock
Image Recognition
Detects facial geometry from live camera data
Claude / ChatGPT
LLM
Generates text by predicting likely next tokens
Credit scoring
Predictive Model
Estimates default risk from financial history
TikTok For You Page
Recommendation
Rapid behavioral feedback loop — every second watched counts
💬
How LLMs Generate Text
Token prediction visualized
Input prompt:
"The capital of France is"
Model scores likely next tokens:
Paris
92%
Lyon
4%
a city
3%
Other
1%
→ Model outputs: Paris — then repeats this for every next token
✅
Self-Check Quiz
Click an answer to check your understanding
Q1 of 4
How does a Large Language Model generate text?
A
It searches a database of correct answers
B
It predicts the most likely next token based on training patterns
C
It copies sentences from its training data
D
It uses logic rules programmed by engineers
✓ LLMs predict statistically likely next tokens — they don't retrieve facts, they generate plausible continuations.
✗ LLMs predict what token is most likely to come next, based on patterns learned during training — not database lookups.
Q2 of 4
What kind of AI powers Netflix's "Because you watched..." recommendations?
A
Image recognition
B
Large Language Model
C
Recommendation engine
D
Reinforcement learning
✓ Netflix uses recommendation engines that match you to content based on behavioral similarity with other users.
✗ Netflix's recommendations come from a recommendation engine — it finds users with similar taste profiles and suggests what they liked.
Q3 of 4
In image recognition, what do the earliest layers detect?
A
Faces, then objects, then colors
B
Edges and textures, then shapes, then objects
C
Objects first, then pixels
D
Colors, then patterns, then labels
✓ Image recognition builds understanding from low-level features (edges) up to high-level concepts (objects).
✗ Early layers detect edges and textures, middle layers detect shapes, later layers detect full objects.
Q4 of 4
Can you swap a recommendation engine for an LLM to do the same job?
A
Yes — all AI systems are interchangeable
B
Yes — LLMs are more powerful so they can do anything
C
No — different AI types are built for different problems and are not interchangeable
D
Only if the LLM is fine-tuned first
✓ Each AI type is purpose-built. An LLM isn't a better recommendation engine — it's a different tool entirely.
✗ Different AI types solve fundamentally different problems with entirely different architectures and use cases.
🧩
Exercises & Worksheets
Apply what you learned
1

Build Your Own Ecosystem Map

List 10 apps or services you use regularly. For each one, identify which AI type powers it (LLM, image recognition, recommendation, predictive). If unsure, make your best guess and explain your reasoning.

🗺 Mapping
2

Test an LLM's Limits

Open any chatbot and ask it a question requiring real-time information (e.g., today's stock price). Observe how it responds. What does this reveal about how LLMs work vs. search engines?

🔍 Experiment
3

Reverse-Engineer a Recommendation

Look at your current Netflix or Spotify recommendations. Pick 3 items and write: What past behavior triggered this? What does the algorithm 'think' you like? Do you agree?

📝 Reflection
4

Explain the Token Bar Chart

Using the LLM token prediction diagram as inspiration, create your own example. Pick a sentence fragment, list 4 possible next words with rough probability estimates, and explain why the model would rank them that way.

🎨 Creative
← Module 02: How ML WorksModule 04: Data →