How can few-shot learning enhance LLM performance?

Question

Options

• ABy enhancing the model's computational efficiency
• BBy providing a large training set to improve generalization
• CBy reducing overfitting through regularization techniques
• DBy offering input-output pairs that exemplify the desired behavior

• A
  4% (1)
• B
  13% (3)
• C
  4% (1)
• D
  79% (19)

Explanation

Few-shot learning works by including a small number of input-output examples directly in the prompt, showing the model exactly what kind of response is expected - this is why D is correct. These examples don't change the model's weights; they simply guide the model's in-context reasoning toward the desired format and behavior.

A is wrong because few-shot prompting has no effect on computational efficiency - it actually increases token usage by adding examples to the prompt. B is wrong because few-shot learning uses only a handful of examples (typically 1–10), not a large training set; it also doesn't involve any training at all. C is wrong because regularization is a training-time technique that modifies how a model learns from data, whereas few-shot prompting happens entirely at inference time.

Memory tip: Think of few-shot as "showing by example" - like a teacher writing three solved problems on the board before asking students to solve a fourth. The key word in D is exemplify: you're giving the model behavioral examples, not changing how it was built.

No community discussion yet for this question.