✅ 50 Solved MCQs on Generative AI 1. What does "generative" mean in the term Generative AI? A) Predicting outcomes B) Analyzing ...
✅ 50 Solved MCQs on Generative AI
1. What does "generative" mean in the term Generative AI?
A) Predicting outcomes
B) Analyzing text
C) Generating new data
D) Recognizing patterns
✅ Answer: C
2. Which of the following is a popular generative model?
A) CNN
B) GAN
C) RNN
D) LSTM
✅ Answer: B
3. Who introduced Generative Adversarial Networks (GANs)?
A) Yann LeCun
B) Andrew Ng
C) Ian Goodfellow
D) Geoffrey Hinton
✅ Answer: C
4. What is the main objective of the generator in a GAN?
A) To classify images
B) To minimize loss
C) To generate fake data
D) To optimize reward
✅ Answer: C
5. What does the discriminator in a GAN do?
A) Generates data
B) Detects fake or real data
C) Optimizes latent variables
D) Adds noise
✅ Answer: B
6. Which loss function is commonly used in GANs?
A) Cross entropy
B) Hinge loss
C) Mean squared error
D) Binary cross entropy
✅ Answer: D
7. What does a Variational Autoencoder (VAE) generate?
A) Only text
B) Deterministic outputs
C) Probabilistic outputs
D) Labels
✅ Answer: C
8. In a VAE, the encoder maps inputs to a:
A) Single point
B) Label
C) Latent space distribution
D) Hidden layer
✅ Answer: C
9. Which generative model uses a score-based approach and noise schedule?
A) GAN
B) VAE
C) Diffusion Model
D) Transformer
✅ Answer: C
10. What is the role of the denoising process in diffusion models?
A) Speed up training
B) Generate sharper images
C) Recover original data from noise
D) Apply attention
✅ Answer: C
11. Which architecture powers ChatGPT and similar models?
A) CNN
B) RNN
C) Transformer
D) Autoencoder
✅ Answer: C
12. What is the key component of the Transformer model?
A) Pooling layer
B) Convolution
C) Attention mechanism
D) ReLU activation
✅ Answer: C
13. What is "text-to-image" generation an example of?
A) Classification
B) Multimodal generation
C) Segmentation
D) Reinforcement
✅ Answer: B
14. What is a diffusion model especially good at?
A) Noise reduction
B) Supervised classification
C) High-fidelity image generation
D) Regression tasks
✅ Answer: C
15. Which of the following is a well-known text-to-image model?
A) GPT-4
B) DALL·E
C) BERT
D) ResNet
✅ Answer: B
16. Which of the following is a text generation model?
A) DALL·E
B) StyleGAN
C) GPT
D) YOLO
✅ Answer: C
17. Which metric measures the quality of generated images?
A) BLEU
B) Inception Score
C) MAE
D) F1-score
✅ Answer: B
18. What is "mode collapse" in GANs?
A) Generator stops training
B) Generator produces limited variety
C) Discriminator wins
D) Data overfitting
✅ Answer: B
19. What does “latent space” mean in generative models?
A) Memory used during training
B) Input feature vector
C) Compressed representation of data
D) Output layer
✅ Answer: C
20. What is one major challenge in training GANs?
A) Too much data
B) Mode diversity
C) Training instability
D) Linear growth
✅ Answer: C
21. In a VAE, what regularizes the latent space?
A) Softmax
B) KL-divergence
C) Dropout
D) Batch norm
✅ Answer: B
22. GPT is pre-trained using which objective?
A) Masked language modeling
B) Next sentence prediction
C) Causal language modeling
D) Text classification
✅ Answer: C
23. Which of the following is not a generative model?
A) VAE
B) GAN
C) ResNet
D) Diffusion Model
✅ Answer: C
24. What type of data can generative AI produce?
A) Only text
B) Text and images
C) Only audio
D) Only numbers
✅ Answer: B
25. Which method can generate synthetic human faces?
A) YOLO
B) StyleGAN
C) LSTM
D) FastText
✅ Answer: B
26. What type of learning is typically used in generative models?
A) Supervised
B) Reinforcement
C) Unsupervised or Self-supervised
D) Manual labeling
✅ Answer: C
27. What is the goal of generative AI in art?
A) Classify images
B) Generate novel artworks
C) Detect plagiarism
D) Enhance compression
✅ Answer: B
28. What ethical concern is common in generative AI?
A) Low accuracy
B) Data leaks
C) Deepfakes
D) Poor optimization
✅ Answer: C
29. Which company created DALL·E and GPT models?
A) DeepMind
B) OpenAI
C) Google
D) Meta
✅ Answer: B
30. Which generative model is best suited for realistic image synthesis?
A) GAN
B) CNN
C) BERT
D) RNN
✅ Answer: A
31. What is an application of generative AI in medicine?
A) Diagnosis prediction
B) Data synthesis for rare conditions
C) Drug classification
D) Signal decoding
✅ Answer: B
32. Which model is used for music generation?
A) MuseNet
B) YOLO
C) InceptionNet
D) DeepSpeech
✅ Answer: A
33. Which technique helps avoid overfitting in generative models?
A) Batch normalization
B) Mode collapse
C) KL divergence
D) Dropout
✅ Answer: D
34. What is "zero-shot" generation?
A) Pretrained on zero data
B) Generating data without specific training on that task
C) Generating 0 images
D) Fine-tuned models
✅ Answer: B
35. What is a synthetic dataset?
A) Collected from users
B) Simulated data generated by models
C) Compressed real data
D) Encrypted data
✅ Answer: B
36. What does GPT stand for?
A) General Purpose Transformer
B) Generative Pretrained Transformer
C) General Pretrained Tuner
D) Global Prediction Transformer
✅ Answer: B
37. What is one benefit of generative AI in education?
A) Replacing teachers
B) Personalized content creation
C) Standard testing
D) Attendance tracking
✅ Answer: B
38. Which of these is a limitation of current generative models?
A) No real-world use
B) Creativity
C) Bias in data
D) Image classification
✅ Answer: C
39. GAN training can be seen as a:
A) Regression task
B) Reinforcement learning task
C) Minimax game
D) Supervised loop
✅ Answer: C
40. Generative AI can be dangerous when used for:
A) Game development
B) Language learning
C) Deepfakes or misinformation
D) Translation
✅ Answer: C
41. Which type of GAN can generate high-resolution images?
A) DCGAN
B) StyleGAN2
C) CycleGAN
D) Pix2Pix
✅ Answer: B
42. What is “prompt engineering” in generative AI?
A) Modifying neural weights
B) Designing model architecture
C) Crafting effective input instructions
D) Tuning loss functions
✅ Answer: C
43. Tokenization in GPT helps with:
A) Detecting punctuation
B) Pretraining the encoder
C) Converting text to numerical inputs
D) Style transfer
✅ Answer: C
44. What is the main challenge in multimodal generation?
A) Too many outputs
B) Aligning different data types
C) Memory issues
D) Lack of activation functions
✅ Answer: B
45. Which model architecture enables image-to-image translation?
A) Pix2Pix
B) GPT
C) ResNet
D) CNN
✅ Answer: A
46. What is the main objective of a text-to-image model like DALL·E?
A) Classify text
B) Generate captions
C) Generate images based on textual description
D) Translate languages
✅ Answer: C
47. What does CLIP model do?
A) Combines language and vision for understanding
B) Crops images
C) Tokenizes input
D) Compresses video
✅ Answer: A
48. What kind of AI generates entirely new and creative content?
A) Reactive AI
B) Weak AI
C) Generative AI
D) Narrow AI
✅ Answer: C
49. Which model translates a sketch to a photo-like image?
A) StyleGAN
B) CycleGAN
C) Sketch2Photo
D) YOLOv4
✅ Answer: C
50. In diffusion models, noise is added in:
A) Reverse order
B) Fixed patterns
C) A series of forward steps
D) Real-time
✅ Answer: C
COMMENTS