Quality and safety: QA and guardrails
Day 5 of 30 · 30 Days of AI
Build a simple QA habit for every AI output
Learning goal
- Create a 5-point QA checklist.
- Apply it to two AI outputs.
Why it matters
- Even good prompts can produce errors or bias.
- QA reduces risk before sharing.
Explanation
- Check facts, numbers, dates.
- Check tone and audience fit.
- Ask for sources/disclaimers when needed.
- Spot sensitive data; redact if required.
Examples
- QA prompt: “Review the draft for accuracy; list uncertainties; suggest fixes.”
- Guardrail: “Do not fabricate data; if unsure, say what’s missing.”
Guided exercise (10–15 min)
- Write a 5-point QA checklist (facts, tone, sources, risks, sensitive data).
- Apply it to one output from Day 4.
Independent exercise (5–10 min)
Apply the checklist to a second output; add a guardrail instruction and compare results.
Self-check
- QA checklist exists.
- Two outputs reviewed and improved.
- Guardrail instruction tested.
Optional deepening
- OpenAI Safety Best Practices: https://platform.openai.com/docs/guides/safety-best-practices