Ethics Advanced Quiz 2
Select your answers and check your results. Use Reset to start again.
Search
Practice Pronunciation (Merriam-Webster)
Navigation
AI Fundamentals Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
AI Fundamentals Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
AI Fundamentals Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Machine Learning Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Machine Learning Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Machine Learning Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Expert Systems Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5,
Expert Systems Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Expert Systems Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Deep Learning Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Deep Learning Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Deep Learning Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Generative AI Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Generative AI Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Generative AI Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Ethics Beginner Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Ethics Intermediate Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Ethics Advanced Quizzes
Quiz 1,
Quiz 2,
Quiz 3,
Quiz 4,
Quiz 5
Advanced Quiz 2
1. What is adversarial robustness testing?
Checking if the AI can compile its own code
Measuring only how fast the model runs
Evaluating how the model behaves on deliberately perturbed or malicious inputs
2. How can red-teaming help surface safety issues in powerful AI systems?
By having experts actively search for failure modes, exploits, and unsafe outputs
By disabling all logging
By only checking spelling in prompts
3. What is scalable oversight?
Oversight that only works for very small models
Methods that let humans effectively supervise increasingly capable systems without reviewing every detail
A way to avoid any human feedback
4. How can weak-to-strong generalization support scalable oversight?
By training only on simple arithmetic tasks
By removing all labels from training data
By using weaker models or humans to supervise stronger models through clever training setups
5. What is reinforcement learning from human feedback (RLHF)?
A method where human preferences guide the reward model used for training
A way to remove humans from the training loop
A technique for compressing training data
6. What is a limitation of RLHF for alignment?
It only applies to text models
It may shape surface behavior without guaranteeing deep, robust alignment of goals
It always produces identical models regardless of data
7. What does reward modeling try to achieve in AI safety?
Minimizing the number of training steps
Maximizing GPU temperature
Learning a function that better represents human judgments about good or bad outcomes
8. How can AI systems be trained to express uncertainty?
By encouraging calibrated confidence scores or refusal behaviors when unsure
By removing all probability outputs
By never providing any answer at all
9. What is impact regularization in AI safety?
A way to speed up matrix multiplication
A training approach that penalizes large, unintended changes to the environment
A technique for formatting log files
10. What is corrigibility in the context of AI?
The ability of AI to always correct human errors
A measure of training dataset size
The tendency of AI systems to accept human oversight, modification, and shutdown without resisting
Previous
Check Quiz
Reset
Next
Other
Timer
00:00
Start
Stop
Reset
Vocabulary Quiz
Score: 0
Reset Score
Submit Answer
Next Word
Spin the Wheel
SPIN
Promo's
Explore More
C# Documentation
C# Tutorials