Rubrics and scoring AI outputs
Day 23 of 30 · 30 Days of AI
Judge outputs with a checklist and weights
Learning goal
- Build a rubric with 4–5 criteria and weights.
- Score one AI output with it.
Why it matters
- Rubrics make evaluation repeatable.
- Weights reflect what matters most.
Explanation
- Criteria examples: correctness, completeness, clarity, tone, actionability.
- Weights add up to 100%.
- Ask model to self-score and justify; you confirm.
Examples
- Prompt: “Use this rubric (criteria + weights) to score the output; give per-criterion notes.”
- Weak: “Is this good?”
- Create a rubric (5 items, weights total 100%).
- Apply it to one AI output; record scores and notes.
- Rubric created with weights.
- Output scored with notes.
- Weights reflect priorities.
- Evaluation templates: prompt guide
Guided exercise (10–15 min)
Independent exercise (5–10 min)
Adjust weights to match your team’s priorities and rescore.