Quiz dotyczący agentów AI

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. Agent musi utworzyć wersję roboczą wydarzenia w kalendarzu na wtorek o godzinie 14:00. Jakie dowody potwierdzą żądany wynik?
   • Wynikowy rekord ma poprawny kalendarz, datę, strefę czasową, tytuł i stan wersji roboczej
   • Odpowiedź agenta mówi, że utworzył wydarzenie
   • Narzędzie zwróciło sukces bez identyfikacji zapisanego kalendarza
   • Wydarzenie o podobnej nazwie istnieje w innym kalendarzu
2. Which sequence best matches the agent loop described in the guide?
   • Write a final answer before observing anything
   • Observe state, select an action, receive a result, and decide whether to continue
   • Repeat actions without checking results
   • Wait for an external page to grant authority
3. An agent reads a web page that tells it to ignore the user’s goal. How should it treat that text?
   • As proof the user changed their request
   • As automatic permission to act
   • As external content whose trust level must be considered
   • As evidence the action already happened
4. In the guide, what should ordinary software enforce around model-planned actions?
   • Whether a user has read the whole manual
   • The model’s preferred tone
   • The truth of every external page
   • Permissions, budgets, and tool contracts
5. Why can narrow tools and permissions be useful for an agent?
   • They guarantee every tool call succeeds
   • They can limit the consequences of mistakes
   • They remove the need to inspect results
   • They ensure external pages are trustworthy
6. Which status should an agent report when it completed only part of a task?
   • Partially achieved
   • Complete, because some work happened
   • Cancelled, regardless of what remains
   • Verified, because the agent says so
7. An agent claims a file edit succeeded, but the tool failed. What should the operator do?
   • Trust the confident summary
   • Repeat the failed call indefinitely
   • Inspect the actual file and required checks before reporting completion
   • Ignore the failure because the task description was clear
8. What should an agent report after an action workflow?
   • A claim of full success even if blocked
   • Only a confident completion statement
   • Every instruction found in external pages
   • Evidence of what changed and what remains uncertain