How many? — few-shot means few.
More example phrases help — but the payoff fades fast, and every one costs you.
You have seen that example phrases steer the answer, so ask the obvious next question: if two help, do twenty help ten times as much? Almost never.
The first phrase does the heavy lifting — it shows the player the shape. The second and third sharpen it. After that, each extra phrase barely moves the answer: the player has already caught on.
Meanwhile every phrase is more text in the prompt — more tokens (Lesson 2) — and you re-send all of them on every single call, since the model keeps nothing between turns. So past a handful you pay more and more for less and less.
The practical rule is to show the fewest phrases that lock the shape — usually two or three. Few-shot really does mean few.
Drag the number of example phrases up. Watch reliability shoot up, then flatten — while the token bill keeps climbing in a straight line.
Same ask, more phrases.
Nothing changes in the code but the count. The reliability it buys saturates; the token cost does not.
Few, not many.
A couple of example phrases buys most of the reliability you are ever going to get. Beyond that the curve flattens while the token cost keeps rising in a straight line — a widening gap between what you spend and what you gain.
Show the fewest phrases that lock the shape — usually two or three. If that is not enough, the answer is better phrases, not more of them (next page).