Lesson 1: The Big Picture
The analogy
Think of a language model as a giant book of numbers.
To write each new word (technically, each token), the computer has to read through the entire book once. It does this again for the next word, and again for the one after that.
That single idea gives you the two questions that decide whether a model will work for you:
| Question | What it really means | Measured in |
|---|---|---|
| 📚 Does the book fit on your shelf? | Is there enough memory to hold the model? | Gigabytes (GB) |
| 👀 How fast can you read the book? | How quickly can the hardware move data out of memory? | Gigabytes per second (GB/s) |
If the book doesn’t fit, nothing else matters: the model won’t load, or it will spill into slow memory and crawl.
If it fits, your reading speed decides whether the model feels instant or painfully slow.
Why not processing power?
Most people assume running AI is about raw computing power. For generating text, it mostly isn’t. The calculations for each word are fairly simple; the bottleneck is fetching billions of numbers from memory, over and over.
That’s why the rest of this course focuses on memory size and memory speed.
💡 There’s one exception, covered in Lesson 6: reading a long prompt or document before answering does depend on computing power, and that’s where GPUs shine.
What’s coming
- Lessons 2–4 answer “Does it fit?”
- Lesson 5 answers “How fast?”
- Lessons 6–8 turn both into decisions.