Books
Facts & News
Courses
Agents
Robotics

About
Contact

Courses › How AI Reads › Part 2

1. Words are not the unit
2. Split a sentence yourself
3. The language tax
4. Check yourself
5. What a context window is
6. Fill the window
7. Long isn't the same as read
8. Check yourself
9. Weights, not a diary
10. Memory features, honestly
11. Final check

🔒

Part 2 opens after the part 1 check

Finish the check at the end of part 1 to open this lesson.

Back to the course

Lesson 8 of 11 · Part 2 · Check yourself · 2 min Draft

Check yourself

Four quick questions on context windows, their cost, and facts lost in the middle.

Four quick questions on Part 2. Look back at the lessons whenever you like.

Check yourself

Get every answer right to open part 3.

1. What is a context window? A setting that hides old messages The model's saved diary of past chats Everything the model can take in at once The box on screen where you type

The context window is the amount of text, in tokens, the model can consider at once. Outside it, nothing exists for the model.

2. Why can each turn of a very long chat cost more? The whole chat is sent, and paid for, again Long chats switch to a pricier model Every reply is saved forever

Services charge per token on every request. A long chat or document re-sent each turn is paid for each turn.

3. A long chat keeps going wrong. What does Hello, AI suggest? Argue until it agrees Repeat yourself in capital letters Paste the whole chat in twice Start a fresh chat and restate the task

Every confused turn stays in the window and feeds the next answer. A fresh chat swaps a messy window for a clean one.

4. As a chat grows, what builds up in the computer's memory? Extra copies of the model Notes the model keeps on each past token Pictures of your screen

Each token leaves notes, the KV cache, so later words can look back. One long chat can fill much of a chip's memory.

← Long isn't the same as read Weights, not a diary →

Books

The Model Race
All books
Errata & corrections

Learn

Facts & News
Courses

Build

Agents
Robotics

Lab

About

Company

Contact