TwiddleLab / Artificial Intelligence / Generative AI: Making New Things

Generating vs Classifying

Lesson 1 of 11 · 5 min · Free, no account needed

What you will learn

• A classifier only needs the border between kinds: it says what a thing is.
• A generative model learns the cloud where real examples live, the data distribution.
• Sampling a random point from that cloud gives a brand new example that still looks real.

Start the lesson

How the lesson goes

1. A classifier tells you what something is. How could a machine make something new?

   Each dot is a real critter, placed by its ear length and body roundness.

2. Be the classifier: split Fluffers from Spikers with one line.

   Drag the two round handles. Fluffers go on the violet side.

3. Now be the generator: make 3 new Fluffers.

   Tap a spot on the map. It turns into a critter with those ears and that body.

4. A machine does the same: learn the cloud, then sample from it.

   Press Learn the cloud, then Sample 6.

5. A generative model learns where the data lives.

   Your 3 critters are the blue dots. The machine's are yellow. Both came from the same cloud.

6. A Fluffer generator draws one new critter. Where does it most likely land?

   Tap A, B or C.

   Sampling picks points in proportion to how common they are in the learned cloud. The middle of the cloud is the most common place.

7. The classifier says all 8 are Fluffers. Find the 3 that are not like any real one.

   Tap a critter: the map shows where it came from. 3 misses allowed.

Next lesson: Autoencoders: Squeeze and Rebuild · Part of the course: Generative AI: Making New Things

Lessons are interactive and need JavaScript. Turn it on to play them.