Where each shows up

You can now look at almost any AI product and name the paradigm behind it. This closing tour does exactly that — six familiar systems, each labeled with the kind of learning that powers it. Notice how the choice always traces back to Chapter 1's question: what feedback was available?

Read the feedback, name the paradigm

The pattern is consistent — the available feedback dictates the tool:

• Had labeled answers? → supervised. Spam filters, medical imaging, credit scoring, translation — anywhere you can gather examples of the right output.
• Only raw data, no labels? → unsupervised. Customer segments, topic discovery, anomaly and fraud detection, and the embeddings behind search and recommendations.
• Could only score outcomes, over time? → reinforcement. Game AI, robotics, and the RLHF that aligns chatbots.

And the biggest systems, as we saw, mix all three.

That's the course

Strip away the jargon and it was one question all along — what feedback does the learner get?

• Labeled answers → supervised.
• No labels, just structure → unsupervised.
• A reward from trial and error → reinforcement.
• And in practice, whatever combination fits the problem.

Next time someone name-drops "reinforcement learning" or "self-supervised pretraining," you'll know exactly what kind of feedback is doing the teaching.

If you enjoyed this, the other courses show these paradigms at work — how ChatGPT learns language, how diffusion models generate images, and how self-driving cars learn from a fleet.