CONTEXT AS THE SCARCE RESOURCE

CHECK YOUR UNDERSTANDING • Why does a model need the harness to resend important facts on every call? ◦ Answer: The model remembers nothing between calls, so anything it should know must be placed in the context again each time. • In the operating-system analogy, what plays the role of the operating system? ◦ Answer: The harness, and especially its context builder, which decides what is loaded into the window. • Why must space be reserved for the model's answer? ◦ Answer: The answer is measured in tokens and shares the same window as the input, so if the input fills the window, the reply is cut off or fails. • Name the three decisions a harness can make about a candidate item. ◦ Answer: Include it in full, summarize it, or exclude it. • Why is retrieving two paragraphs better than pasting a 300-page manual for a simple question? ◦ Answer: It saves cost and time, leaves room for other information, and reduces distraction, so the answer is usually faster and more accurate.