Learn AI. Do. Grow.
← All modules

Basic · module 34 of 100

Context window and memory

Understand how much the model “contains” in a single conversation and what that means.

After this lesson

You know when to repeat information and when a new conversation is better.

In view · How it connects

The worksheet has an edge

  1. Instructions, messages and provided documents take up space.

  2. Only a limited amount of material fits in the window at one time.

  3. Consciously re-include important agreements when the context becomes large.

Remember: The context window and any stored memory are different things. Open het volledige schema ↗

Limited memory

The model has a context window. Everything outside that window is “forgotten”. Long conversations can lose earlier details.

What to do

Repeat important facts if you continue. Or start a new conversation for a new task.

Practical tip

Put the core data at the top or in a fixed prompt. Occasionally ask for a summary of the agreements so far.

New practice example

From explanation to a usable result.

A conversation has limited space. Saved preferences and the current conversation context are different things.

Fictional practice situation

In a long conversation about a fictitious sports day, the location changes from hall A to hall B. The final start time has not yet been confirmed. You are starting a new conversation.

View a possible solution

This is how it can be done

Provide a short handover: location hall B, replaces hall A; start time unknown; goal is a participant message. Add the last confirmed note as the source and leave out old drafts.

Why this works

A new conversation does not have those agreements by default. A summary may contain errors; compare the handover with the original decisions.

Make it slightly more difficult

Add two contradictory notes about the start hour. Show how you preserve the conflict without making a choice yourself.

Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.

Your workplace · learning by doing

Make something. See what changes.

Ende, your light pink practice buddy

The fictitious source is included. Only when selecting ‘Let Ende answer’ do you send your assignment to the AI service. Do not use private data.

Checking the AI connection… You can already view the written example.

Your result

No answer yet

Present your assignment to Ende.
Read the answer, change one thing, and compare.

You remain in control

What do you want to improve in your next attempt?

My own check

On this device only. AI answers remain only in this page memory. This is self-review, not a grade.

Why this approach works

A new conversation does not have those agreements by default. A summary may contain errors; compare the handover with the original decisions.

A step further

Add two contradictory notes about the start hour. Show how you preserve the conflict without making a choice yourself.

Try & discover · no obligation

Fill the context window

Make your input longer. Then reserve more space for the answer. Where is the limit?

Divide the space into practice tokens
2800 of 4000 tokens used.

Blue: inputGreen: answer spaceGrey: free space

1200tokens free
€0.0088fictitious costs at full usage

The specified input and reserved output fit together in this window.

Practice model, not a real tokenizer or product limit. Input also includes instructions and history. Fictitious rate: € 0,002 per 1.000 input tokens and € 0,006 per 1.000 output tokens. Actual rates and calculation rules differ.

This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.

A prompt to get you started

Here are the agreements so far: [summary]. Continue with the next task: [task]. Use only these agreements.

Apply it

After a long conversation, your instruction refers to agreements from much earlier. The assistant misses an important limitation.

Which approach best fits this situation?

  • Assume that every old message always remains available.
  • Provide a controlled summary of the relevant agreements again.
  • Let the assistant reconstruct the forgotten agreement itself.

With JavaScript, you can check your approach and receive immediate feedback.

Test your insight

What happens to information outside the context window?

  • The model always remembers everything.
  • The model no longer has that information available in that conversation.
  • It is automatically saved.
Extra practice

Describe a situation in which it is better to start a new conversation instead of continuing in a long conversation.

After a long brainstorm about ideas, I start a new conversation to flesh out one chosen idea. This keeps the context clean.

A question? I'll help you on your way.

Drag Aldo to move it. Use the arrow keys on Aldo with the keyboard, Enter to save the position and Escape to cancel.