Track · 2:30 · Liner note
Context Window Seat
A context window is the maximum amount of text, measured in tokens, that a language model can consider at once, covering the instructions, the conversation so far and the reply.
Every language model reads through a window of limited size, measured in tokens, that has to hold instructions, history, retrieved documents and the reply itself. Context Window Seat is about getting a good place in it. What sits in the window shapes the answer, and what does not fit is invisible to the model.
Spend the space deliberately. Put in the instructions and the relevant rows, leave out the rest and summarize old turns instead of keeping them whole. A bigger window is room, not a reason to fill it.