Teach the model by example. Go from giving no examples (zero-shot) to one (one-shot) to several (few-shot), and feel how each added example locks the output into the exact shape you want.
Why: zero-shot means you give the instruction and no examples — fastest to write, and modern models are good at it. When: reach for it first for common, well-understood tasks. When it fails: the format drifts or the model misreads an unusual label.
Classify the sentiment of this review as positive, negative,
or neutral.
Review: "The battery dies in an hour but the screen is gorgeous."Why: a single example pins down the exact output format and labelling you expect. When: use it when zero-shot gets the right idea but the wrong shape (extra words, wrong casing, explanations you did not ask for).
Classify the sentiment as positive, negative, or neutral.
Review: "Setup was painless and it just works."
Sentiment: positive
Review: "The battery dies in an hour but the screen is gorgeous."
Sentiment:Why: several examples teach an edge case a single one cannot — here, that a mixed review should be "neutral". When: use 3-5 examples for tasks with tricky boundaries or a custom category scheme the model has never seen.
Classify the sentiment as positive, negative, or neutral.
Review: "Setup was painless and it just works."
Sentiment: positive
Review: "Stopped charging after a week. Avoid."
Sentiment: negative
Review: "Great camera, but the app is a buggy mess."
Sentiment: neutral
Review: "The battery dies in an hour but the screen is gorgeous."
Sentiment:Why: the model copies everything about your examples — spacing, casing, punctuation. When examples disagree (one says "Positive", another "positive"), the model picks randomly. Where: make every example identical in format, and cover each label at least once so none looks rare.
# Do: identical format, balanced labels
Review: "..." Sentiment: positive
Review: "..." Sentiment: negative
Review: "..." Sentiment: neutral
# Avoid: mixed casing, missing a category, ragged spacing
Review:"..." -> Positive
review: "..." sentiment = negative