People choose the more helpful and safe answer from two responses.
Move model behavior closer to the desired preference.
Choose two correct statements.
Human rankings can guide adaptation, but evaluator criteria, bias, consent, confidentiality, and post-update safety require governance and reevaluation.
Detailed explanation
New output quality and safety must be tested.
The feedback can inform adaptation.
Feedback data is also a quality and governance concern.
Evaluator perspective and data bias can be inherited.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01 Domain 2.1の基盤モデル適応、人のフィードバック、責任あるAIを確認する。
Official references
What to confirm
人の評価データの利点と、評価者・データ品質の限界を説明できる。
Key points
- Preference
- Evaluator
- Reevaluation
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Compare the definition with the AIF-C01 objective and the service documentation before applying it.
- Keep the concept separate from a particular answer choice or implementation step.
Foundation review
Human feedback
People's comparisons or judgments about model outputs used for learning or evaluation.
Preference data
Records indicating which of multiple outputs a reviewer prefers and why.