Lesson 01Loading progress
From rules to learning
AI did not begin with chatbots. Start with a simple question: can we write down every rule?
Listen Experiment
7 minChecking progress…
Chapter 1
Why rules gave way to learning in many tasks, and how language models became assistants.
AI did not begin with chatbots. Start with a simple question: can we write down every rule?
Give a tiny model examples. Then test whether it learned something useful beyond them.
What if the examples were sentences, and the missing label was already in the text?
Predicting text is a foundation. Becoming useful in a conversation takes additional work.
Chapter 2
Slow down the journey from your prompt to a generated response.
You type a message. An answer appears. Let's slow down the part in between.
A model doesn't see a sentence the way you do. Let's meet its building blocks.
One small prediction. Then another. Here's how a response grows.
Chapter 3
Where the model’s abilities come from.
A model is not a box of stored replies. Start with two numbers you can change yourself.
A prediction misses. Turn that error into a direction for the next update.
An ID identifies a token. A vector gives the model numbers it can work with.
Chapter 4
Connect words, relationships, and the transformer.
Same words. Different order. A very different story.
Watch one token gather information from the tokens around it.
Put tokens, attention, and learned transformations into one picture.
Chapter 5
Turn a language model into a useful conversational partner.
A model that can continue text is not automatically a helpful assistant.
Two answers can sound good. Which one deserves to be reinforced?
The same scores can produce different choices. Turn the dial and see why.
Chapter 6
Know when to check, when to retrieve, and when to use tools.
A polished sentence can contain a supported fact and an invented detail.
What a model can see now differs from what an app saves for later.
Give an assistant evidence and a calculator. Inspect the path to its answer.
Your final challenge: turn a confident draft into an answer you can defend.