Yapay Zeka Ürün Yönetimi Sınavı

Testi çözdüğünüzde cevaplar ve açıklamalar görünür.

1. What should a team identify before selecting an AI model for a feature?
   • The user task, current alternative, constraints, and successful outcome
   • The vendor’s largest benchmark
   • A preferred architecture regardless of need
   • A public launch date
2. Why can a high model score fail to show that a feature helps users?
   • High scores are always fabricated
   • Product metrics cannot be measured
   • Models cannot be evaluated before launch
   • Model metrics do not directly measure task completion or user effort
3. A support assistant closes more tickets, but customers reopen them. What should the team examine?
   • Only the number of closed tickets
   • Whether the interface uses a chatbot
   • Resolution and repeat contact alongside closure count
   • The model’s parameter count
4. What should the product define for unsupported requests and uncertain outputs?
   • A promise that the model will answer everything
   • Acceptable behavior, fallbacks, and ways to reach another route
   • A rule to conceal uncertainty
   • A longer product name
5. Why record model and prompt versions during release evaluation?
   • To prove every output is reproducible word for word
   • To avoid user research
   • To make future monitoring unnecessary
   • To investigate changes and regressions against the artifact actually used
6. When should a team consider a simpler non-AI approach?
   • When it can solve the user problem with less complexity
   • Never; product value requires AI
   • Only after every model is tested
   • When a model score exceeds a threshold
7. How should a team evaluate a new AI product change?
   • Use only a demonstration prompt
   • Use representative users and tasks, then measure product outcomes and failures
   • Count conversations regardless of completion
   • Compare different audiences without accounting for assignment
8. Which user recovery feature fits a product with uncertain outputs?
   • Remove the ability to edit
   • Hide the model’s uncertainty
   • Let users correct mistakes, cancel work, or choose another route
   • Require users to repeat the entire task