Prompt engineering — the few-shot examples do most of the work
How you phrase and exemplify a task often changes accuracy more than which model you pick, and the discipline is measuring that rather than trusting intuition about wording.
Not yet scoped.
Why
The PoC runs the same task under zero-shot, few-shot and formatted prompts against a fixed eval set, turning prompt choices into a measured comparison.
How it works
Not yet built.
Where it lands in Jayverse
Auditor / gitboard: run agent-driven checks under a fixed eval set before trusting them. Compare zero-shot, few-shot and formatted prompts for authority audits and invariant reviews rather than tuning wording by intuition.
alice-tech: apply the same measured comparison to the daily tech-report summarization prompt. A fixed eval set catches drift before it silently degrades what gets logged.
Key expressions
Words and phrases from this page worth keeping, with the Korean meaning and the sentence they come from.
Expression
뜻 · 쓰이는 자리
do most of the work
대부분의 역할을 하다·핵심 요인이다 · 결과를 좌우하는 가장 큰 요인을 가리킬 때. "the few-shot examples do most of the work"
trust intuition about
~에 대한 직감을 그냥 믿다 · 검증 없이 감각적 판단에 의존하는 것. "rather than trusting intuition about wording"
a fixed eval set
고정된 평가 데이터셋 · 여러 방식을 같은 기준으로 비교할 때 쓰는 세트. "against a fixed eval set"
a measured comparison
실측으로 비교한 것(감이 아니라 수치로) · 직관 대신 데이터로 판단하는 방식. "turning prompt choices into a measured comparison"
zero-shot / few-shot
예시 없이 / 몇 개의 예시로 하는 프롬프트 방식 · 모델에게 과제를 지시하는 서로 다른 방법. "under zero-shot, few-shot and formatted prompts"