Page 54 - CyberSecurityEssentialsEnglish

Term Common abbreviation Plain-language explanation

Tuning a trained model using human
Reinforcement
comparisons of candidate answers, so
learning from RLHF
that it follows instructions and keeps a
human feedback
helpful tone.

Review Questions
1. What is the difference between artificial intelligence as a field and an AI system as a

product or service?
2. Why does AlphaGo’s success not mean that the system possesses general human

intelligence?
3. How do symbolic systems differ from machine learning, and why might a hybrid system
be appropriate?
4. Explain the difference among classification, numerical forecasting, ranking, generation,

and optimization.
5. What is the difference among training, inference, and evaluation?

6. Why do data quality and collection methods affect a system’s output?
7. Why might a system’s performance decline when the data or operating environment
changes? Give an example.

8. How does a language model construct a sentence step by step, and why does that process
not guarantee factual accuracy?
9. Why might AI succeed at a complex task and fail at one that appears simpler?

10. What is the difference among acceleration, augmentation, and automation?
11. What makes human review meaningful rather than merely formal?
12. How would you build a local test set for a task you perform regularly?

13. Why might an AI system be excellent at producing a text draft yet unsuitable for making
a financial or health decision without review?
14. What kinds of task does AI usually do well, and what condition must hold before you rely

on the output?

53