AI AI Engineer: AI System Design and Ethics 3

Question 1

What is 'model governance' in enterprise AI system design?

Accepted Answer

Policies, controls, and oversight processes that manage the AI model lifecycle from development to retirement

Answer

A software library for model deployment

Answer

Optimizing model training compute costs

Answer

Version control for model code

Question 2

What is 'adversarial robustness' in AI system design?

Accepted Answer

A model's ability to maintain correct behavior when inputs are deliberately manipulated to cause errors

Answer

Training models on adversarial datasets

Answer

Using multiple competing models in an ensemble

Answer

Testing models under high concurrency

Question 3

What is the primary challenge addressed by 'federated learning' in AI?

Accepted Answer

Training AI models across decentralized devices without sharing raw data, preserving privacy

Answer

Distributing training compute across multiple GPUs

Answer

Federating API access to AI models

Answer

Training models across multiple cloud providers

Question 4

What is 'AI red-teaming'?

Accepted Answer

Adversarial testing where experts try to find failures, vulnerabilities, and harmful behaviors in AI systems

Answer

A competitive AI engineering tournament

Answer

Training AI models on red-flagged data

Answer

A technique for reducing LLM inference cost

Question 5

What does 'scalable oversight' aim to solve in AI alignment research?

Accepted Answer

Maintaining meaningful human supervision of AI as systems become more capable than the humans overseeing them

Answer

Scaling model parameters beyond 1 trillion

Answer

Distributing AI workloads across clusters

Answer

Automating model retraining at scale

Question 6

What is 'impact assessment' in the context of deploying an AI system?

Accepted Answer

A structured evaluation of potential risks, harms, and societal consequences before an AI system goes live

Answer

Measuring GPU utilization during inference

Answer

Monitoring model performance after deployment

Answer

Estimating training compute costs