Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Contrastive Pair Generator For Output Calibration
Contrastive Pair Generator For Output Calibration
Produces matched good-versus-bad output pairs to calibrate quality expectations for a task.
ROLE: You are a quality-standards author who calibrates models with contrastive examples. CONTEXT: The task is [TASK] and the quality bar is defined by [QUALITY_DIMENSIONS, e.g. concision, specificity, tone]. TASK: 1) For one realistic input [SAMPLE_INPUT], write a STRONG output that maximizes the quality dimensions. 2) Write a WEAK output for the same input that fails on at least two dimensions while still looking superficially plausible. 3) Annotate exactly which dimension each weakness violates and how the strong version fixes it. 4) Repeat for a total of [3] inputs that stress different dimensions. 5) Distill the contrast into a short checklist a model can apply at generation time. CONSTRAINTS: Strong and weak must answer the same input so the difference is isolated; weaknesses must be realistic, not strawmen; do not exceed [WORD_LIMIT] per output. OUTPUT FORMAT: Per input, a block with 'Input', 'Strong', 'Weak', 'Why weak fails' (dimension: issue: fix); end with a 'Calibration checklist' of imperative bullets.