The Context Window

A conversation with a language model can feel continuous. The model experiences no such continuity.

Every time a model generates a response, it starts from scratch. There is no memory of previous conversations, no accumulated relationship, no growing familiarity. There is only what is in front of it right now: a window of text, fixed in size, containing everything it is allowed to see.

What fills that window shapes everything. The conversation so far, instructions, documents you've shared, background context. When the window fills up, something gets trimmed. The model does not know what is missing. It works with what it has.

Understanding this changes how you use these tools. The model is not a person who remembers you. It is a process that reads a document and continues it. That document has edges, and the edges matter.