14 Sep 100 Generative AI MCQ (Multiple Choice Questions) with Answers

Posted at 21:50h in mcq by

 

1) What is the primary goal of Generative AI?
  1. To classify data into predefined categories.
  2. To create new, original content based on learned patterns.
  3. To optimize a specific business process.
  4. To predict a numerical value based on input features.
Show Answer
Answer: b
Explanation
Generative AI learns the underlying patterns/distribution of the training data so it can produce new, original content, rather than only classifying or predicting.

2) Which of the following is a classic example of a generative model?

  1. Logistic Regression
  2. Support Vector Machine (SVM)
  3. Naive Bayes Classifier
  4. Variational Autoencoder (VAE)
Show Answer
Answer: d
Explanation
A Variational Autoencoder is a generative model that learns a latent representation of the data and can sample from it to create new data. The others are discriminative models.

3) What is the fundamental difference between discriminative and generative models?

  1. Discriminative models are always supervised, while generative models are unsupervised.
  2. Discriminative models learn the boundary between classes (P(y|x)), while generative models learn the distribution of the data (P(x,y) or P(x)).
  3. Generative models are less computationally intensive than discriminative models.
  4. Discriminative models can create new data, while generative models can only classify.
Show Answer
Answer: b
Explanation
Discriminative models model the decision boundary P(y|x), while generative models model how the data itself is distributed (P(x) or P(x,y)) — which lets them create new samples.

4) Which architecture is the foundation for most modern Large Language Models (LLMs) like GPT?

  1. Recurrent Neural Network (RNN)
  2. Long Short-Term Memory (LSTM)
  3. Transformer
  4. Convolutional Neural Network (CNN)
Show Answer
Answer: c
Explanation
Modern LLMs such as the GPT series are built on the Transformer architecture and its self-attention mechanism.

5) What does the “GPT” in GPT-4 stand for?

  1. General Purpose Transformer
  2. Generative Pre-trained Transformer
  3. Graph Processing Tool
  4. Generative Predictive Text
Show Answer
Answer: b
Explanation
GPT stands for Generative Pre-trained Transformer — a generative model, pre-trained on large text corpora, using the Transformer architecture.

6) In the context of Generative AI, what is a “prompt”?

  1. A type of error message.
  2. The output generated by the AI model.
  3. The input text or instruction given to an AI model to guide its output.
  4. A training dataset for the model.
Show Answer
Answer: c
Explanation
A prompt is the input text or instruction supplied to a generative model to steer what it produces.

7) What is “zero-shot learning” in the context of LLMs?

  1. The model performs a task without any training data for that specific task.
  2. The model learns from zero examples.
  3. The model’s performance is zero on a new task.
  4. The model requires zero computational resources.
Show Answer
Answer: a
Explanation
Zero-shot means the model tackles a task it was never explicitly trained on, relying only on the instruction in the prompt.

8) What is “few-shot learning”?

  1. The model is trained for a few epochs only.
  2. The model performs a task after being given a small number of examples in the prompt.
  3. The model can only perform a few tasks.
  4. The model uses a few parameters.
Show Answer
Answer: b
Explanation
Few-shot prompting supplies a handful of worked examples inside the prompt so the model can infer the desired pattern.

9) What is a “hallucination” in Generative AI?

  1. When the model generates content that is visually impressive.
  2. When the model generates content that is plausible but factually incorrect or nonsensical.
  3. When the model takes a long time to generate a response.
  4. When the model refuses to answer a question.
Show Answer
Answer: b
Explanation
A hallucination is fluent, confident output that is factually wrong or invented — a well-known limitation of LLMs.

10) What is the purpose of a “tokenizer” in an LLM?

  1. To encrypt the input text.
  2. To break down text into smaller units (tokens) that the model can process.
  3. To generate the final output text.
  4. To evaluate the quality of the generated text.
Show Answer
Answer: b
Explanation
A tokenizer splits text into tokens (words, sub-words or characters) and maps them to numeric IDs the model can consume.

11) Which of the following is NOT a major family of generative models?

  1. Generative Adversarial Networks (GANs)
  2. Variational Autoencoders (VAEs)
  3. Transformer-based models
  4. Random Forest
Show Answer
Answer: d
Explanation
Random Forest is an ensemble of decision trees used for classification/regression — a discriminative method, not a generative model family.

12) In a Generative Adversarial Network (GAN), what are the two main components?

  1. Encoder and Decoder
  2. Generator and Discriminator
  3. Input and Output layers
  4. Transformer and Attention
Show Answer
Answer: b
Explanation
A GAN pits a Generator (which creates fake samples) against a Discriminator (which tries to tell real from fake).

13) What is the role of the “Generator” in a GAN?

  1. To distinguish between real and fake data.
  2. To create new data samples that mimic the real data.
  3. To classify the input data.
  4. To optimize the loss function.
Show Answer
Answer: b
Explanation
The Generator takes random noise and transforms it into samples intended to look like the real training data.

14) What is the role of the “Discriminator” in a GAN?

  1. To create new data samples.
  2. To distinguish between real and fake (generated) data.
  3. To encode the input data.
  4. To decode the latent space.
Show Answer
Answer: b
Explanation
The Discriminator is a classifier that judges whether a given sample came from the real dataset or from the Generator.

15) What is a key characteristic of Variational Autoencoders (VAEs)?

  1. They use a game-theoretic approach with two competing networks.
  2. They learn a probabilistic, continuous latent space.
  3. They are primarily used for text generation.
  4. They are a type of discriminative model.
Show Answer
Answer: b
Explanation
VAEs encode inputs into a probabilistic (Gaussian) latent distribution, which allows smooth interpolation and sampling of new data.

16) Which model type is known for generating highly realistic images but can suffer from “mode collapse”?

  1. Variational Autoencoder (VAE)
  2. Generative Adversarial Network (GAN)
  3. Transformer
  4. Diffusion Model
Show Answer
Answer: b
Explanation
GANs produce sharp, realistic images but are notoriously unstable to train and prone to mode collapse.

17) What is “mode collapse” in the context of GANs?

  1. When the generator produces a limited variety of outputs, failing to capture the full diversity of the training data.
  2. When the discriminator becomes too good at its job.
  3. When the training process is too slow.
  4. When the model generates a single, perfect image.
Show Answer
Answer: a
Explanation
In mode collapse the Generator finds a few outputs that fool the Discriminator and keeps producing only those, losing diversity.

18) How do Diffusion Models generate data?

  1. By using a generator and discriminator.
  2. By learning to reverse a gradual noising process.
  3. By encoding data into a latent space and decoding it.
  4. By using a recurrent network to predict the next pixel.
Show Answer
Answer: b
Explanation
Diffusion models progressively add noise to data during training and then learn to denoise step by step, generating samples from pure noise.

19) Which of the following models is an example of a Diffusion Model?

  1. GPT-4
  2. BERT
  3. DALL-E 2
  4. StyleGAN
Show Answer
Answer: c
Explanation
DALL-E 2 uses a diffusion-based image generation pipeline. GPT-4 and BERT are language models and StyleGAN is a GAN.

20) What is “latent space”?

  1. The final output layer of a neural network.
  2. A compressed, lower-dimensional representation of the data learned by a model.
  3. The raw input data.
  4. The space where prompts are written.
Show Answer
Answer: b
Explanation
Latent space is the compressed internal representation where similar data points sit close together, enabling sampling and interpolation.

21) What is the primary function of the “self-attention” mechanism in a Transformer?

  1. To process data sequentially, one token at a time.
  2. To allow the model to weigh the importance of different tokens in the input sequence relative to each other.
  3. To reduce the dimensionality of the input data.
  4. To generate the final output probabilities.
Show Answer
Answer: b
Explanation
Self-attention computes relationships between all tokens in a sequence, letting each token attend to the most relevant others.

22) What is a “positional encoding” in a Transformer model?

  1. A technique to encode the position of each token in the sequence, since the model itself is order-agnostic.
  2. A method to encrypt the input text.
  3. A way to initialize the model’s weights.
  4. A type of activation function.
Show Answer
Answer: a
Explanation
Because self-attention has no built-in notion of order, positional encodings are added to token embeddings to inject sequence position information.

23) Which company developed the GPT series of models?

  1. Google
  2. Meta
  3. OpenAI
  4. Microsoft
Show Answer
Answer: c
Explanation
OpenAI developed the GPT family, including GPT-3, GPT-4 and the ChatGPT products built on them.

24) Which model is known for its bidirectional understanding of text and is often used for tasks like sentiment analysis?

  1. GPT
  2. BERT
  3. GAN
  4. VAE
Show Answer
Answer: b
Explanation
BERT is an encoder-only Transformer that reads context in both directions, making it strong at understanding tasks like classification and sentiment analysis.

25) What does “multimodal” mean in the context of Generative AI?

  1. The model can only generate one type of data (e.g., text).
  2. The model can understand and generate multiple types of data (e.g., text, images, audio).
  3. The model has multiple layers.
  4. The model is trained on multiple datasets of the same type.
Show Answer
Answer: b
Explanation
Multimodal models work across more than one modality — for example taking text and images as input and producing text, images or audio as output.

26) What is a “foundation model”?

  1. A small, specialized model for a single task.
  2. A large, pre-trained model that can be adapted (fine-tuned) for a wide range of downstream tasks.
  3. The first layer of a neural network.
  4. A model that is not yet trained.
Show Answer
Answer: b
Explanation
Foundation models are trained at massive scale on broad data and then adapted to many downstream tasks through fine-tuning or prompting.

27) What is “fine-tuning”?

  1. The initial, large-scale training of a model on a massive dataset.
  2. The process of further training a pre-trained model on a smaller, task-specific dataset.
  3. Adjusting the learning rate during training.
  4. A type of data augmentation.
Show Answer
Answer: b
Explanation
Fine-tuning continues training of an already pre-trained model on a smaller, domain- or task-specific dataset to specialise its behaviour.

28) What is “prompt engineering”?

  1. The process of building the hardware for an AI model.
  2. The practice of designing and refining prompts to get the desired output from a generative model.
  3. Writing the code for the model’s architecture.
  4. Cleaning and preparing the training data.
Show Answer
Answer: b
Explanation
Prompt engineering is the craft of wording, structuring and iterating prompts so a generative model returns the output you actually want.

29) What is “Reinforcement Learning from Human Feedback” (RLHF)?

  1. A technique to train a model using a reward signal derived from human preferences.
  2. A method for unsupervised learning.
  3. A way to generate data without any human involvement.
  4. A type of model architecture.
Show Answer
Answer: a
Explanation
RLHF trains a reward model from human comparisons of outputs, then optimises the language model against that reward signal.

30) What is the purpose of RLHF in models like ChatGPT?

  1. To make the model faster.
  2. To align the model’s outputs with human values, making them more helpful, harmless, and honest.
  3. To increase the model’s parameter count.
  4. To reduce the cost of training.
Show Answer
Answer: b
Explanation
RLHF is used for alignment — steering the model toward responses humans judge as helpful, harmless and honest.

31) What is a “temperature” parameter in text generation?

  1. A parameter that controls the randomness of the model’s output. Higher values lead to more creative but less predictable text.
  2. A parameter that controls the speed of generation.
  3. A parameter that sets the length of the output.
  4. A parameter that determines the model’s accuracy.
Show Answer
Answer: a
Explanation
Temperature scales the sampling distribution: low values make output focused and deterministic, high values make it more random and creative.

32) What does “top-k sampling” do?

  1. It always picks the most probable next word.
  2. It restricts the sampling pool to the top ‘k’ most likely next words.
  3. It picks a word randomly from the entire vocabulary.
  4. It sorts the output words alphabetically.
Show Answer
Answer: b
Explanation
Top-k sampling truncates the probability distribution to the k most likely tokens and samples only from those.

33) What is the main challenge of training very large Generative AI models?

  1. They are too easy to train.
  2. They require immense computational resources, data, and energy.
  3. They are not capable of generating complex content.
  4. They can only be trained by one company.
Show Answer
Answer: b
Explanation
Training frontier models demands enormous compute (thousands of GPUs), massive curated datasets and significant energy and cost.

34) What is “overfitting” in the context of generative models?

  1. The model generates too much data.
  2. The model learns the training data too well, including its noise and specific details, and fails to generalize to new, unseen data.
  3. The model is too simple to learn the data patterns.
  4. The model trains too quickly.
Show Answer
Answer: b
Explanation
Overfitting means the model memorises training specifics (including noise) instead of learning generalisable patterns, so it performs poorly on new data.

35) What is “underfitting”?

  1. The model is too complex for the data.
  2. The model has not learned the underlying patterns of the data well enough.
  3. The model generates too little data.
  4. The model trains for too long.
Show Answer
Answer: b
Explanation
Underfitting occurs when a model is too simple or trained too little to capture the true structure of the data, performing poorly even on training data.

36) Which of the following is a common application of Generative AI in art?

  1. Creating realistic 3D models from text descriptions.
  2. Colorizing black and white photos.
  3. Generating original paintings in the style of a specific artist.
  4. All of the above.
Show Answer
Answer: d
Explanation
All three are established generative-AI art applications: text-to-3D, image colorization and style-based image synthesis.

37) What is “style transfer”?

  1. Transferring a file from one computer to another.
  2. Applying the artistic style of one image to the content of another.
  3. Changing the format of an image.
  4. Training a model on a new dataset.
Show Answer
Answer: b
Explanation
Style transfer re-renders the content of a source image using the visual style (colour, texture, brushwork) of a reference image.

38) How is Generative AI used in drug discovery?

  1. To generate new molecular structures that could be potential drugs.
  2. To create marketing materials for pharmaceutical companies.
  3. To diagnose diseases from medical images.
  4. To manage patient records.
Show Answer
Answer: a
Explanation
Generative models propose novel molecular structures with desired properties, greatly speeding up candidate discovery and screening.

39) What is “synthetic data”?

  1. Data that is collected from the real world.
  2. Data that is artificially generated by a computer model, not from real-world events.
  3. Data that is encrypted.
  4. Data that is stored in the cloud.
Show Answer
Answer: b
Explanation
Synthetic data is produced by algorithms/models rather than being captured from real-world events, and is designed to mimic real data statistically.

40) What is a potential benefit of using synthetic data for training AI models?

  1. It is always more accurate than real data.
  2. It can help overcome privacy concerns and data scarcity issues.
  3. It is easier to collect than real data.
  4. It requires no validation.
Show Answer
Answer: b
Explanation
Synthetic data avoids exposing personal information and can fill gaps where real, labelled data is scarce or expensive.

41) What is a “deepfake”?

  1. A type of deep neural network.
  2. A synthetic media in which a person in an existing image or video is replaced with someone else’s likeness.
  3. A very accurate prediction made by an AI.
  4. A philosophical concept about AI consciousness.
Show Answer
Answer: b
Explanation
A deepfake is AI-generated or manipulated audio/video in which a person’s face, voice or likeness is swapped or fabricated.

42) What is the primary ethical concern associated with deepfakes?

  1. They are too expensive to create.
  2. They can be used to spread misinformation, create non-consensual pornography, and commit fraud.
  3. They are not realistic enough to be believable.
  4. They require too much data to create.
Show Answer
Answer: b
Explanation
Deepfakes enable disinformation, non-consensual intimate imagery, impersonation and fraud — hence the strong ethical and legal concern.

43) What is “AI alignment”?

  1. The process of aligning the text in a document.
  2. The research area focused on ensuring AI systems act in accordance with human values and intentions.
  3. The process of connecting multiple AI models together.
  4. A type of data preprocessing.
Show Answer
Answer: b
Explanation
AI alignment research seeks to make AI systems reliably pursue the goals and values humans actually intend.

44) What does the term “black box” refer to in AI?

  1. A physical box that contains the AI’s hardware.
  2. A model whose internal workings are not easily understood or interpretable by humans.
  3. A model that is trained on data from a black box.
  4. A type of error in AI models.
Show Answer
Answer: b
Explanation
“Black box” describes models whose internal decision-making is opaque, making their outputs hard to explain or audit.

45) What is “explainable AI” (XAI)?

  1. A field of research that aims to make AI models’ decisions more transparent and understandable.
  2. A type of AI that can explain jokes.
  3. An AI that is used for educational purposes.
  4. A marketing term for AI.
Show Answer
Answer: a
Explanation
Explainable AI develops methods and tools that make a model’s reasoning transparent and interpretable to humans.

46) What is the risk of “bias” in Generative AI models?

  1. The models can generate content that is offensive, stereotypical, or discriminatory, reflecting biases present in their training data.
  2. The models may be too slow.
  3. The models may require too much power.
  4. The models may not be creative enough.
Show Answer
Answer: a
Explanation
Models reproduce and can amplify the stereotypes and imbalances present in their training corpora, producing harmful or discriminatory output.

47) Where does bias in a Generative AI model primarily come from?

  1. The programmers’ personal opinions.
  2. The large-scale datasets used for pre-training, which often contain societal biases.
  3. The hardware used to train the model.
  4. The users’ prompts.
Show Answer
Answer: b
Explanation
The primary source is the pre-training data, which reflects historical and societal biases present in text and images scraped from the web.

48) What is “model stealing” or “model extraction”?

  1. Physically stealing the server that hosts an AI model.
  2. An attack where an adversary tries to replicate the functionality of a proprietary model by querying it.
  3. A technique to compress a large model into a smaller one.
  4. A method for transferring a model from one cloud to another.
Show Answer
Answer: b
Explanation
Model extraction uses large numbers of API queries to train a clone that approximates the victim model, potentially exposing its value and vulnerabilities.

49) What is a “prompt injection” attack?

  1. Injecting a virus into the AI’s code.
  2. A technique where a user provides a malicious prompt designed to override the model’s original instructions or safety guardrails.
  3. A way to make the model generate faster.
  4. A method for fine-tuning a model.
Show Answer
Answer: b
Explanation
Prompt injection smuggles instructions into the model’s context (often via untrusted content) to hijack its behaviour and bypass safety rules.

50) What is “data leakage” in the context of LLMs?

  1. When the model’s training data is accidentally exposed in its outputs.
  2. When the model’s data is stolen by hackers.
  3. When the model uses too much memory.
  4. When the training data is lost.
Show Answer
Answer: a
Explanation
Data leakage occurs when a model regurgitates memorised training content — including private or sensitive information — in its responses.

51) Which of the following is a popular open-source text-to-image model?

  1. GPT-4
  2. Stable Diffusion
  3. BERT
  4. LaMDA
Show Answer
Answer: b
Explanation
Stable Diffusion is an openly released latent diffusion text-to-image model. GPT-4, BERT and LaMDA are language models.

52) What is “in-painting” in image generation?

  1. The process of coloring an image.
  2. The process of filling in a missing or masked part of an image.
  3. The process of increasing the resolution of an image.
  4. The process of converting a 2D image to 3D.
Show Answer
Answer: b
Explanation
In-painting fills a masked region of an image with newly generated content that is consistent with its surroundings.

53) What is “out-painting”?

  1. The process of extending an image beyond its original borders.
  2. The process of removing an object from an image.
  3. The process of converting an image to black and white.
  4. The process of adding text to an image.
Show Answer
Answer: a
Explanation
Out-painting (uncropping) generates new pixels around an existing image to enlarge the canvas or change its aspect ratio.

54) What does “text-to-image” generation refer to?

  1. Generating an image based on a textual description (prompt).
  2. Generating text that describes an image.
  3. Converting an image into a text file.
  4. Creating a text document with image placeholders.
Show Answer
Answer: a
Explanation
Text-to-image models take a natural-language prompt and synthesise a matching image.

55) What is “text-to-video” generation?

  1. Creating a video from a script.
  2. Generating a video clip based on a textual prompt.
  3. Adding subtitles to a video.
  4. Transcribing the audio from a video.
Show Answer
Answer: b
Explanation
Text-to-video models generate a moving video clip directly from a written prompt.

56) What is “text-to-3D” generation?

  1. Creating a 3D model from a text description.
  2. Writing a review for a 3D movie.
  3. Converting a 3D model to a text file.
  4. Creating a 2D image from a 3D model.
Show Answer
Answer: a
Explanation
Text-to-3D systems synthesise a three-dimensional model, mesh or scene from a textual description.

57) What is “text-to-speech” (TTS) synthesis?

  1. Converting a text document into a speech audio file.
  2. Transcribing a speech recording into text.
  3. Translating text from one language to another.
  4. Analyzing the sentiment of a text.
Show Answer
Answer: a
Explanation
TTS converts written text into spoken audio. The reverse process (speech to text) is automatic speech recognition.

58) What is “voice cloning”?

  1. Creating a new voice from scratch.
  2. Using AI to create a synthetic copy of a person’s voice.
  3. Changing the pitch of a voice.
  4. Recording a voice in a soundproof room.
Show Answer
Answer: b
Explanation
Voice cloning learns the characteristics of a speaker’s voice from samples so the model can make them “say” new text.

59) Which of the following is a common metric for evaluating the quality of generated text?

  1. BLEU (Bilingual Evaluation Understudy)
  2. PSNR (Peak Signal-to-Noise Ratio)
  3. FID (Fréchet Inception Distance)
  4. Accuracy
Show Answer
Answer: a
Explanation
BLEU compares generated text with reference text using n-gram overlap, and is widely used for machine translation and text generation. PSNR and FID are image metrics.

60) Which metric is commonly used to evaluate the quality of generated images?

  1. BLEU
  2. ROUGE
  3. FID (Fréchet Inception Distance)
  4. Perplexity
Show Answer
Answer: c
Explanation
FID measures the distance between feature distributions of real and generated images; lower FID indicates more realistic images.

61) What does the “perplexity” metric measure in language modeling?

  1. How confused the model is by the data. A lower perplexity indicates a better model.
  2. The speed at which the model generates text.
  3. The number of parameters in the model.
  4. The grammatical correctness of the output.
Show Answer
Answer: a
Explanation
Perplexity is derived from the model’s probability assigned to the data — the lower the perplexity, the better the model predicts the text.

62) What is “Inception Score” (IS) used for?

  1. Evaluating the quality and diversity of generated images.
  2. Measuring the speed of an AI model.
  3. Evaluating the sentiment of text.
  4. Calculating the error rate of a classifier.
Show Answer
Answer: a
Explanation
The Inception Score uses a pre-trained classifier to judge how recognisable (quality) and varied (diversity) a set of generated images is.

63) What is a “hyperparameter”?

  1. A parameter that is learned during the training process.
  2. A parameter that is set before the training process begins (e.g., learning rate, number of layers).
  3. A parameter that is too large for the model.
  4. A parameter that is related to the hardware.
Show Answer
Answer: b
Explanation
Hyperparameters are configured by the practitioner before training (learning rate, batch size, layer count) rather than learned from data.

64) What is a “learning rate”?

  1. The speed at which the model processes data.
  2. A hyperparameter that controls how much the model’s weights are updated during training.
  3. The rate at which the model generates new content.
  4. The accuracy of the model.
Show Answer
Answer: b
Explanation
The learning rate scales the size of each gradient-descent weight update — too high causes divergence, too low causes very slow training.

65) What is “batch size”?

  1. The total size of the training dataset.
  2. The number of training examples used in one iteration to update the model’s weights.
  3. The size of the model’s memory.
  4. The number of layers in the model.
Show Answer
Answer: b
Explanation
Batch size is how many samples are processed before one parameter update. It affects memory use, speed and gradient stability.

66) What is a “gradient”?

  1. The final output of the model.
  2. A vector that points in the direction of the steepest increase of a function, used in optimization algorithms like gradient descent.
  3. A type of neural network layer.
  4. A measure of the model’s accuracy.
Show Answer
Answer: b
Explanation
The gradient is the vector of partial derivatives of the loss. Optimisers move the weights in the opposite (downhill) direction.

67) What is “backpropagation”?

  1. The process of propagating errors backward through the network to update weights.
  2. The process of feeding data forward through the network.
  3. A type of data augmentation.
  4. A method for initializing weights.
Show Answer
Answer: a
Explanation
Backpropagation applies the chain rule to send error gradients from the output layer back through the network so weights can be updated.

68) What is a “loss function”?

  1. A function that measures the difference between the model’s predictions and the actual target values.
  2. A function that generates new data.
  3. A function that initializes the model’s weights.
  4. A function that loads the data.
Show Answer
Answer: a
Explanation
The loss (cost) function quantifies prediction error, and training aims to minimise it.

69) What is the “Adam” optimizer?

  1. A popular optimization algorithm used for training deep learning models.
  2. A type of neural network architecture.
  3. A dataset for training generative models.
  4. A company that develops AI models.
Show Answer
Answer: a
Explanation
Adam (Adaptive Moment Estimation) combines momentum with adaptive per-parameter learning rates and is a standard deep-learning optimiser.

70) What is “dropout”?

  1. A regularization technique where randomly selected neurons are ignored during training to prevent overfitting.
  2. A technique to reduce the size of the dataset.
  3. A method for generating new data.
  4. An error in the training process.
Show Answer
Answer: a
Explanation
Dropout randomly deactivates neurons during training, forcing the network to learn redundant, more generalisable representations.

71) What is “transfer learning”?

  1. Transferring data from one computer to another.
  2. Using knowledge gained from one task to improve performance on a different but related task.
  3. A type of reinforcement learning.
  4. A method for data compression.
Show Answer
Answer: b
Explanation
Transfer learning reuses features learned on a source task to boost learning on a related target task, usually saving data and compute.

72) What is “zero-shot” image classification?

  1. Classifying images into categories the model has never seen before, often using a textual description of the class.
  2. Classifying an image with zero pixels.
  3. A model that has zero accuracy.
  4. Classifying an image without any computational cost.
Show Answer
Answer: a
Explanation
Zero-shot classification recognises unseen classes by matching image features against text descriptions of those classes (as in CLIP).

73) Which model, developed by Google, is known for its ability to generate highly realistic images from text prompts?

  1. Imagen
  2. BERT
  3. GPT-3
  4. Word2Vec
Show Answer
Answer: a
Explanation
Imagen is Google’s text-to-image diffusion model. BERT and GPT-3 are text models, and Word2Vec is an older word-embedding technique.

74) What is “Midjourney”?

  1. A research lab focused on AI safety.
  2. A popular proprietary AI image generator.
  3. A type of GAN.
  4. A dataset of images.
Show Answer
Answer: b
Explanation
Midjourney is a proprietary text-to-image generation service known for its distinctive artistic output.

75) What is the core idea behind “contrastive learning”?

  1. To train a model by contrasting positive pairs (similar items) with negative pairs (dissimilar items).
  2. To train a model on a very large dataset.
  3. To use a very high learning rate.
  4. To train a model without any labels.
Show Answer
Answer: a
Explanation
Contrastive learning pulls representations of matched pairs together and pushes unmatched pairs apart, without needing class labels.

76) Which model architecture is used by CLIP (Contrastive Language-Image Pre-training)?

  1. A GAN
  2. A VAE
  3. A combination of a text encoder and an image encoder.
  4. A simple feedforward network.
Show Answer
Answer: c
Explanation
CLIP trains a text encoder and an image encoder jointly so that matching image–caption pairs land close together in a shared embedding space.

77) What is the main advantage of using a Transformer over an RNN for text generation?

  1. Transformers are much smaller.
  2. Transformers can process the entire sequence in parallel, leading to faster training and better handling of long-range dependencies.
  3. Transformers are easier to implement.
  4. Transformers require less data.
Show Answer
Answer: b
Explanation
Self-attention lets Transformers process all positions in parallel and model long-range dependencies directly, unlike sequential RNNs.

78) What is “attention” in the context of neural networks?

  1. A mechanism that allows the model to focus on the most relevant parts of the input when producing an output.
  2. A technique to make the model pay attention to the user.
  3. A method for visualizing the model’s internal state.
  4. A type of activation function.
Show Answer
Answer: a
Explanation
Attention assigns learned weights to different parts of the input, letting the model dynamically prioritise the most relevant information.

79) What is a “multi-head” attention mechanism?

  1. An attention mechanism with multiple sets of query, key, and value weight matrices, allowing the model to focus on different aspects of the input simultaneously.
  2. An attention mechanism that can only be used on multi-core processors.
  3. An attention mechanism with a single head.
  4. A type of optimizer.
Show Answer
Answer: a
Explanation
Multi-head attention runs several attention operations in parallel with different learned projections, then concatenates their outputs.

80) What is a “masked language model” (MLM)?

  1. A model that is trained to predict a masked (hidden) word in a sentence based on its context. BERT is a prime example.
  2. A model that wears a mask.
  3. A model that can only generate text about masks.
  4. A model that is trained to predict the next word in a sequence.
Show Answer
Answer: a
Explanation
MLM training hides random tokens and asks the model to recover them using bidirectional context — the core pre-training objective of BERT.

81) What is a “causal language model”?

  1. A model that is trained to predict the next word in a sequence, using only the words that came before it. GPT is a prime example.
  2. A model that can predict the cause of an event.
  3. A model that is used for causal inference.
  4. A model that is not allowed to predict the future.
Show Answer
Answer: a
Explanation
Causal (autoregressive) language models use a left-to-right mask so each token can only attend to preceding tokens — the basis of the GPT family.

82) What is the purpose of a “system prompt” in a conversational AI?

  1. To provide a high-level instruction or context that guides the AI’s overall behavior and persona throughout the conversation.
  2. To prompt the user to ask a question.
  3. To display a system error message.
  4. To start the operating system.
Show Answer
Answer: a
Explanation
The system prompt sets persistent instructions — role, tone, constraints — that shape the model’s behaviour across the whole conversation.

83) What is “AI safety”?

  1. Ensuring the physical safety of AI researchers.
  2. An interdisciplinary field focused on preventing accidents, misuse, or other harmful consequences arising from AI systems.
  3. A type of firewall for AI models.
  4. A method for backing up AI models.
Show Answer
Answer: b
Explanation
AI safety studies how to build and deploy AI systems so they do not cause accidents, harm or misuse.

84) What is the “alignment problem”?

  1. The challenge of ensuring that the goals and behaviors of powerful AI systems are aligned with human values and intentions.
  2. A problem with the alignment of text in a document.
  3. The problem of aligning the layers of a neural network.
  4. A problem with the alignment of the hardware.
Show Answer
Answer: a
Explanation
The alignment problem is the difficulty of specifying and guaranteeing that capable AI systems pursue what humans actually want.

85) What is a “superintelligence”?

  1. An intellect that is much smarter than the best human brains in practically every field.
  2. A very large AI model.
  3. An AI that is very good at a single task.
  4. A type of computer processor.
Show Answer
Answer: a
Explanation
Superintelligence is a hypothetical intellect vastly exceeding human capability across essentially all domains.

86) What is “AGI”?

  1. Artificial General Intelligence: a hypothetical AI with human-level cognitive abilities across a wide range of tasks.
  2. Advanced Generative Intelligence.
  3. Automated Graphic Interface.
  4. Artificial GAN Intelligence.
Show Answer
Answer: a
Explanation
AGI refers to a general-purpose AI that can learn and perform any intellectual task at roughly human level, unlike today’s narrow AI.

87) What is “AI ethics”?

  1. The study of the moral principles and values that should guide the design, development, and use of AI.
  2. A set of laws governing AI.
  3. A marketing term for AI.
  4. A type of AI model.
Show Answer
Answer: a
Explanation
AI ethics examines the moral principles — fairness, accountability, transparency, privacy — that should shape AI development and deployment.

88) What is “algorithmic fairness”?

  1. The idea that AI systems should be designed to avoid creating or reinforcing unfair bias.
  2. The speed at which an algorithm runs.
  3. The fairness of the algorithm’s name.
  4. A type of sorting algorithm.
Show Answer
Answer: a
Explanation
Algorithmic fairness is the goal of designing and auditing systems so they do not produce discriminatory or unjust outcomes.

89) What is “GDPR”?

  1. A type of Generative AI model.
  2. A data protection and privacy regulation in the European Union that has implications for AI.
  3. A company that makes AI chips.
  4. A software library for AI.
Show Answer
Answer: b
Explanation
The General Data Protection Regulation governs how personal data of EU residents may be collected, processed and stored — with direct consequences for AI training and deployment.

90) What is “data privacy” in the context of Generative AI?

  1. The right of individuals to control how their personal data is collected, used, and shared by AI models.
  2. The process of making data public.
  3. A type of data encryption.
  4. A method for data compression.
Show Answer
Answer: a
Explanation
Data privacy concerns an individual’s control over their personal data — including whether it is used to train models or reproduced in outputs.

91) What is a “neural network”?

  1. A network of computers.
  2. A computing system vaguely inspired by the biological neural networks that constitute animal brains.
  3. A type of internet connection.
  4. A network for sharing data.
Show Answer
Answer: b
Explanation
An artificial neural network is composed of layered, interconnected units (neurons) loosely inspired by biological brains.

92) What is a “parameter” in a neural network?

  1. A user setting.
  2. A weight or bias in the network that is learned during training.
  3. The number of layers in the network.
  4. The size of the input data.
Show Answer
Answer: b
Explanation
Parameters are the internal weights and biases adjusted by the optimiser during training — distinct from user-set hyperparameters.

93) What is an “epoch” in machine learning?

  1. One complete pass through the entire training dataset.
  2. A unit of time.
  3. The time it takes to train a model.
  4. The time it takes to generate a single output.
Show Answer
Answer: a
Explanation
One epoch means the model has seen every training example once. Training usually runs for many epochs.

94) What is “data augmentation”?

  1. The process of increasing the amount of data by adding slightly modified copies of existing data.
  2. The process of deleting data.
  3. The process of encrypting data.
  4. The process of visualizing data.
Show Answer
Answer: a
Explanation
Data augmentation creates modified copies of existing samples (rotations, crops, flips, paraphrases) to enlarge the dataset and reduce overfitting.

95) What is “cross-validation”?

  1. A technique for evaluating a model’s performance by partitioning the data into subsets, training on some and testing on others.
  2. A method for validating the data’s integrity.
  3. A way to cross-reference data from different sources.
  4. A type of neural network layer.
Show Answer
Answer: a
Explanation
Cross-validation (e.g., k-fold) rotates which subset is held out for testing, giving a more reliable estimate of generalisation performance.

96) What is a “confusion matrix”?

  1. A table used to evaluate the performance of a classification model.
  2. A matrix that is confused.
  3. A type of generative model.
  4. A method for data processing.
Show Answer
Answer: a
Explanation
A confusion matrix tabulates true positives, true negatives, false positives and false negatives, from which precision, recall and F1 are derived.

97) What is “precision” in the context of classification metrics?

  1. The ratio of correctly predicted positive observations to the total predicted positive observations. (True Positives / (True Positives + False Positives))
  2. The ratio of correctly predicted positive observations to the all observations in the actual class.
  3. The overall accuracy of the model.
  4. The speed of the model.
Show Answer
Answer: a
Explanation
Precision = TP / (TP + FP) — of everything the model labelled positive, how much was actually positive.

98) What is “recall”?

  1. The ratio of correctly predicted positive observations to the all observations in the actual class. (True Positives / (True Positives + False Negatives))
  2. The ratio of correctly predicted positive observations to the total predicted positive observations.
  3. The overall accuracy of the model.
  4. The speed of the model.
Show Answer
Answer: a
Explanation
Recall (sensitivity) = TP / (TP + FN) — of all the genuinely positive cases, how many the model managed to find.

99) What is the “F1-score”?

  1. A single metric that combines both precision and recall into one value, calculated as the harmonic mean of precision and recall.
  2. The score of the first model trained.
  3. A type of loss function.
  4. The score of the final model.
Show Answer
Answer: a
Explanation
F1 = 2 × (precision × recall) / (precision + recall), balancing the two metrics and especially useful on imbalanced datasets.

100) What is the future potential of Generative AI?

  1. It is a niche technology with limited applications.
  2. It is expected to revolutionize fields like content creation, healthcare, and scientific research, but also poses significant ethical and societal challenges.
  3. It will be replaced by discriminative AI.
  4. Its development has already stopped.
Show Answer
Answer: b
Explanation
Generative AI is expected to transform many industries while raising serious concerns around bias, misinformation, privacy, safety and economic disruption.
100 Agentic AI MCQ (Multiple Choice Questions) with Answers
Print page
0 Likes
Studyopedia Editorial Staff
contact@studyopedia.com

We work to create programming tutorials for all.

No Comments

Post A Comment