🚀 Generative AI Interview Questions with Answers — Part 6
🧪 Model Behavior and Safety
51. What are hallucinations in AI models?
Hallucinations occur when an AI model generates false, misleading, or completely fabricated information while sounding confident.
Example: An AI may:
• Invent fake facts
• Generate incorrect citations
• Create nonexistent references
Hallucinations are one of the biggest challenges in Generative AI systems.
52. Why do hallucinations happen?
Hallucinations happen because LLMs predict the most probable next token rather than verifying facts.
Common causes:
• Lack of real-time knowledge
• Insufficient context
• Poor training data
• Ambiguous prompts
• Weak retrieval systems
• Overgeneralization
LLMs generate responses based on patterns, not actual understanding.
53. How can hallucinations be reduced?
Common Techniques:
1. Using RAG systems
2. Better prompting
3. Fine-tuning with high-quality data
4. Fact-checking systems
5. Human review
6. Grounding responses in documents
7. Limiting unsupported generation
Example: Instead of asking:
“Explain this topic.”
Use:
“Answer only using the provided document.”
This improves factual accuracy.
54. What is bias in Generative AI?
Bias refers to unfair, prejudiced, or unbalanced outputs generated by AI models.
Bias may come from:
• Training data
• Human annotations
• Historical inequalities
• Cultural imbalance
Examples:
• Gender bias
• Racial bias
• Political bias
• Language bias
Bias can negatively impact fairness and trustworthiness.
55. How do you detect biased outputs?
Bias can be detected through:
• Human evaluation
• Fairness testing
• Benchmark datasets
• Output audits
• Diversity analysis
• Adversarial testing
Teams often test models using prompts across:
• Different genders
• Ethnicities
• Languages
• Cultures
Responsible AI requires continuous monitoring for bias.
56. What are the ethical concerns in Generative AI?
Major Ethical Concerns:
• Misinformation
• Deepfakes
• Copyright issues
• Privacy violations
• Job displacement
• Harmful content generation
• Bias and discrimination
Organizations developing AI systems must follow ethical and responsible AI practices.
57. What is model alignment?
Model alignment means ensuring AI systems behave according to human values, goals, and safety expectations.
Aligned models aim to be:
• Helpful
• Honest
• Safe
• Reliable
Techniques used:
• RLHF
• Safety tuning
• Content filtering
• Human feedback
Alignment is critical for trustworthy AI systems.
58. What is content filtering?
Content filtering is the process of detecting and blocking harmful, unsafe, or inappropriate AI outputs.
Examples:
• Hate speech filtering
• Violence detection
• Adult content moderation
• Misinformation prevention
Content filtering improves AI safety and user protection.
59. What are guardrails in GenAI systems?
Guardrails are safety mechanisms that control AI behavior and prevent harmful outputs.
Examples:
• Blocking dangerous prompts
• Restricting unsafe actions
• Preventing prompt injection attacks
• Enforcing company policies
Guardrails help ensure safe and responsible AI usage.
60. Why is responsible AI important?
Responsible AI ensures that AI systems are:
• Fair
• Transparent
• Safe
• Ethical
• Accountable
Benefits:
• Builds user trust
• Reduces harmful outcomes
• Improves compliance
• Supports ethical innovation
As Generative AI adoption grows, responsible AI practices are becoming essential for companies like OpenAI, Google DeepMind, and Anthropic.
Double Tap ❤️ For More
