Tokens and Context
Models can only work with the information that fits into their current context.
Models can only work with the information that fits into their current context.
Small pieces of text the model processes.
The working space available for the current request.
Recent and explicit instructions often matter most.
Understand the working model
Long prompts are not automatically better. Context has a capacity, so every instruction, example and pasted document competes for attention. Put the goal and critical constraints where they are easy to find.
When a task uses a long document, ask the model to first identify the relevant section, then perform the task on that section. This reduces noise and makes verification easier.
Practice
Apply the pattern above to one real task you already do. Keep the first version small enough that you can inspect every input and output.
Do not paste everything you have “just in case.” Irrelevant context can make the answer less focused.
Give the model the smallest complete context needed for the task, and make the goal and constraints explicit.