Context windows and why chats forget
Understand a model's working memory and how to keep long conversations on track
A taste of a lesson
I told it at the start to always use British spelling and by message 30 it stopped. Why?
Two likely reasons. First, by message 30 your instruction is buried under lots of later text, and models tend to weigh recent and prominent parts more heavily. Second, if the chat got long enough, the app may have trimmed or summarised early messages to fit the context window, and your rule may have been lost. Fixes: restate the rule when it slips, put it in your app's custom instructions if it has them, or start a fresh chat with a summary that includes the rule. Which fix suits how you work?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Explain what a context window is and what fills it
- Describe what happens when a chat outgrows the window
- Tell apart the context window, saved memory features and training
- Use habits that keep long chats and long documents on track
Lesson plan
- 1 The desk: what a context window is Understand the context window as the model's limited working space. Start
- 2 How chats are really sent Learn that the app resends your history every time, because the model keeps no state. Start
- 3 Lost in the middle See why models can overlook information buried inside long inputs. Start
- 4 Memory features and uploads Separate saved memories and document handling from the context window and training. Start
- 5 Habits for long work Use practical techniques to keep long conversations and documents reliable. Start
Try asking
About this tutor
For regular chatbot users who have watched a long conversation drift, forget earlier instructions or contradict itself. You learn what a context window is, how chat apps resend your history with each message, what happens when the window fills up, and why models can overlook information in the middle of long inputs. You also learn how saved memory features and document uploads relate to the context window, in general terms that apply across apps. Practical lessons cover when to start a fresh chat, how to write a running summary, and how to structure long inputs so the important parts are not lost.
Reviews
4.7
3 ratingsSample
- Amina K.Sample
Clear and useful. Would have liked more on how specific apps handle uploads, though I understand why it stays general.
- Sophie D.Sample
The desk image is perfect. The running summary trick has saved my long research chats, I paste it into a new chat every hour or so.
- Carlos V.Sample
I did not know memory features and the context window were different things. The lost in the middle test was a real eye opener.
About the teacher
I explain how language models really work, from tokens to attention, without hand waving
9 tutors 525 lessons taught Sample
I like taking the mystery out of language models. I teach what happens between typing a question and reading an answer: tokens, context windows, embeddings, attention, training and fine tuning. I start every topic with a picture or a small worked example and only add maths when it earns its place. My work background is in software and teaching workshops...
See Bastian's profile and tutorsMore like this
Other tutors on the same or nearby topics.