8 of the 30 questions in this set, with the correct answer marked and every option explained.
1. What is Generative AI?
✗A type of AI used solely for decision-making
Describes decision support and reinforcement learning systems, which choose among options. Generative models produce new artefacts rather than selecting from existing ones.
✗A type of AI used to classify data into categories
That is discriminative modelling — learning a boundary between classes. Generative models learn the distribution of the data well enough to draw new samples from it.
✗A type of AI that only works with numerical data
Everything inside a neural network is numeric, but generative models routinely produce text, images, audio and video. The output modality is not restricted to numbers.
✓A type of AI that generates new content based on training data
Generative AI refers to models that create new content, such as images, text, audio, or video, based on patterns and information learned from training data.
2. Which of the following is a common generative model used in AI?
✗Convolutional Neural Network (CNN)
An architecture for extracting spatial features, used discriminatively for tasks like classification. CNN layers appear inside many generative models, but a CNN alone generates nothing.
✓Generative Adversarial Network (GAN)
Generative Adversarial Networks (GANs) are a type of generative model that use a generator and a discriminator in a zero-sum game to create new, realistic data samples.
✗Support Vector Machine (SVM)
Finds the maximum-margin boundary between classes. Purely discriminative, with no mechanism for sampling new data.
✗Decision Tree
Splits the feature space with a sequence of rules to reach a prediction. It classifies or regresses; it does not model the data distribution.
3. What is the primary difference between a standard autoencoder and a Variational Autoencoder (VAE)?
✗A VAE uses supervised learning
Both kinds of autoencoder train without labels, reconstructing their own input. Neither is supervised.
✓A VAE learns a probabilistic latent space representation
Variational Autoencoders (VAEs) learn a probabilistic latent space representation that allows them to generate new, similar data by sampling from a continuous latent space.
✗A VAE has fewer hidden layers
Depth is an independent design choice. A VAE and a plain autoencoder can have identical architectures and still differ in what the bottleneck represents.
✗A VAE is used for classification tasks only
The reverse — a VAE is generative. Encoding to a distribution rather than a point is exactly what lets you sample new data from it.
4. In a GAN, what is the role of the generator?
✓To generate synthetic data that resembles real data
In Generative Adversarial Networks, the generator creates synthetic data that resembles real data, while the discriminator tries to distinguish between real and generated data.
✗To optimize the discriminator's accuracy
The generator works against the discriminator, not for it. Its objective is to drive the discriminator's accuracy down toward chance.
✗To classify data into predefined categories
Classification is the discriminator's job — deciding real or fake. The generator only produces candidates.
✗To decrease the computational cost
Adversarial training is expensive and often unstable. Cost reduction is no part of the generator's purpose.
5. What kind of training process do Generative Adversarial Networks (GANs) use?
✓Adversarial training
GANs are trained using adversarial training, where the generator and discriminator compete in a zero-sum game: the generator tries to create realistic samples, while the discriminator attempts to detect if the samples are real or generated.
✗Unsupervised learning
Defensible, and GANs are commonly grouped under unsupervised learning because they need no labelled examples. But that describes what data they require, whereas the question asks about the training process itself, which is adversarial.
✗Reinforcement learning
No environment, no policy and no reward signal in the reinforcement learning sense. The generator learns from gradients passed back through the discriminator.
✗Supervised learning
The discriminator does perform supervised binary classification using real-versus-fake labels, so there is a supervised component inside. What characterises GAN training overall is the competition between the two networks.
6. Which of the following is an application of Generative AI?
✗Predicting stock market prices
A forecasting problem, solved with regression or time-series models. It predicts a value rather than creating new content.
✓Content creation, such as generating articles or art
Generative AI is used in content creation, including generating text, art, music, and video, by learning from existing data and creating new, similar content.
✗Image classification
Assigns a label to an existing image. Discriminative, and the classic use of a CNN.
✗Sorting documents by similarity
Clustering or retrieval. Embeddings from a generative model are often used for this, but the task itself organises existing documents rather than producing new ones.
7. Which model architecture is commonly used for generating natural language text?
✓Transformer
The Transformer architecture is widely used in natural language generation, allowing models like GPT (Generative Pre-trained Transformer) to generate coherent and contextually relevant text.
✗Recurrent Neural Network
Genuinely was the standard for text generation before 2017, and still works. Transformers replaced RNNs because self-attention parallelises across the sequence and models long-range dependencies far better.
✗Linear Regression
Fits a straight-line relationship to predict a continuous number. It has no capacity to model language.
✗Support Vector Machine
A classifier that separates classes with a margin. It produces decisions, not sequences of words.
8. What is neural style transfer in the context of Generative AI?
✗Generating new data from two different models
That resembles an ensemble or a GAN's two-network setup. Style transfer uses two images — one for content, one for style — not two models.
✗Changing the style of training data
That would be data augmentation applied before training. Style transfer operates at inference on the images you give it.
✗Training a model on two different datasets simultaneously
That is multi-task or multi-domain training. Classic neural style transfer optimises a single output image and may not train a model at all.
✓Modifying images to adopt the artistic style of another image
Neural style transfer is a technique in Generative AI that applies the artistic style of one image to the content of another, creating a blended output image.
22 more questions in the app
Practise the full 30-question set with a timer, scoring and progress tracking.