Course 6, lesson 54 of 100, Ages 12+
Context windows
An AI’s working memory
Like I’m 5
When you chat with an AI, it can only look at a certain amount of the conversation at once, like reading through a window. That's its context window.
The big idea
The context window is the maximum number of tokens a model can consider at once: your messages, its replies, and any documents you paste in. Some models handle a few pages, others whole books.
When a conversation is longer than the window, older parts fall out of view, so the model may forget early details. Many chatbots don't learn from your chat afterwards; they only remember within the window, unless a separate memory feature saves notes.
Examples
- Long chats: After a very long chat, it may forget your name from the start.
- Pasting a book: Large windows let you ask questions about a whole report.
- Fresh start: A new chat usually begins with an empty window.
How it works
- Your messages and documents enter the context window.
- The model reads everything in the window to reply.
- If it overflows, the oldest parts drop out of view.
Check your understanding
- What is a context window?
- Options: How much text a model can look at at once; A real window in the computer; A list of passwords.
Answer: How much text a model can look at at once. It's the model's working memory for the current conversation. - Why might a chatbot forget something from early in a long chat?
- Options: It fell outside the context window; It got tired; It deleted the internet.
Answer: It fell outside the context window. Once text leaves the window, the model can't see it.
Remember
A context window is the AI's working memory. Too much text, and early parts fall out.
Talk about it
How much of a conversation do you remember after an hour? How is that like a context window?
Go deeper
Window sizes are measured in tokens. Attention cost grows with sequence length, so long-context models use efficient attention variants. Retrieval (RAG) and memory tools extend effective context.