✅ Top AI Interview Questions with Answers: Part-5 🧠

41. What is tokenization and stemming?
• Tokenization: Splitting text into individual units (words, sentences).
E.g., "I love AI" → ["I", "love", "AI"]
• Stemming: Reducing words to their root form.
E.g., "running", "runner" → "run"
Used in NLP to preprocess and normalize text.

42. Explain BERT and its use cases
BERT (Bidirectional Encoder Representations from Transformers) is a transformer-based model by Google.
• Reads text bidirectionally (context from both sides).
• Pre-trained on a large corpus, fine-tuned for tasks.
Use cases:
• Sentiment analysis
• Question answering
• Named entity recognition
• Text classification

43. What is the role of attention in transformers?
Attention allows models to focus on relevant parts of the input sequence when making predictions.
• Helps capture relationships between words regardless of distance.
• Key component in models like BERT, GPT, T5.
It improves understanding of context and meaning in sequences.

44. What is a language model?
A language model predicts the next word or sequence based on previous context.
Trained on large text data to understand grammar, meaning, and structure.
Examples: GPT, BERT, LLaMA.
Used in chatbots, autocomplete, summarization, translation, etc.

45. Explain YOLO in object detection
YOLO (You Only Look Once) is a real-time object detection system.
• Processes image in one pass (single CNN)
• Outputs bounding boxes and class probabilities
• Fast and efficient — ideal for real-time apps like surveillance, autonomous vehicles.

46. What is Explainable AI (XAI)?
XAI makes AI decisions understandable to humans.
• Helps build trust
• Useful in regulated industries (healthcare, finance)
Techniques include SHAP, LIME, attention maps.

47. What is model interpretability vs explainability?
• Interpretability: How easily humans can understand the model (especially linear or simple models).
• Explainability: Explaining decisions of complex models (e.g., deep learning) using tools or approximations.
Both are key for trust, compliance, and debugging.

48. How do you deploy a machine learning model?
Steps to deploy:
1. Train and validate model
2. Save model (e.g., Pickle, Joblib)
3. Wrap in an API (Flask, FastAPI)
4. Containerize (Docker)
5. Host on server/cloud (AWS, Heroku, Azure)
6. Monitor performance and update regularly

49. What are ethical concerns in AI?
• Bias fairness
• Privacy data security
• Job displacement
• Misinformation deepfakes
• Lack of transparency
Addressed via regulations, audits, responsible AI frameworks.

50. What is prompt engineering in LLMs?
Prompt engineering is crafting inputs to guide large language models like GPT to produce accurate and desired outputs.
• Uses techniques like few-shot, zero-shot, and chain-of-thought prompting
• Critical for building AI apps, chatbots, and tools using LLMs effectively.

💬 Double Tap ♥️ For More