NEW
Font size
WorksheetsGen AI Quiz
Total questions: 20
Worksheet time: 7mins
Who are the primary instructors for the DeepLearning.AI short course titled "ChatGPT Prompt Engineering for Developers"?
Andrew Ng and Isa Fulford
Jay Alammar and Gwendolyn Stripling
Lukasz Kaiser and Illia Polosukhin
Adola Puni and Chinuo
Which of the following tasks is listed as an application area for using LLM APIs in the "ChatGPT Prompt Engineering for Developers" course outline?
Training discriminative machine learning models
Sentiment classification
Computing cross-entropy loss functions
Generating positional encoding vectors
In the context of effective prompt engineering, what key characteristic did a contributor emphasize as being crucial, highlighting that prompt development is usually not perfect on the first try?
It must use a JSON format.
It is beginner-friendly.
It must be iterative.
It must incorporate beam search.
In the Transformer architecture, what purpose does the self-attention layer primarily serve in the encoder?
It prevents the decoder from attending to future positions in the output sequence.
It adds positional encoding vectors to the input embeddings.
It transforms text (e.g., translation, grammar correction).
It allows the model to look at other words in the input sequence for clues, associating the word being processed with relevant context for better encoding.
What is the fundamental difference between a generative AI model and a discriminative machine learning model?
Generative models are generally used for text-to-text tasks, while discriminative models are used for image generation.
Discriminative models learn the joint probability distribution of input and output.
Generative models generate new content (like text, images, or audio), while discriminative models classify or predict labels for existing data points.
Generative models must use labeled data, whereas discriminative models use unlabeled data.
Why is the use of delimiters in prompting advised, according to the sources?
To enforce the use of few-shot examples.
To expand the context automatically.
To prevent the model from getting confused by clearly separating instructions from user input.
To reduce the computation of multi-headed attention.
What is the term for when a Large Language Model generates confident but inaccurate or irrelevant output, often due to being trained on incomplete or contradictory data?
Embedding
Iteration
Hallucination
Softmaxing
Retrieval-Augmented Generation (RAG) is a technique that addresses which primary limitation of foundation models?
The inability to use the Python SDK for API calls.
The model’s inability to access private, proprietary, or real-time data beyond its original knowledge cutoff.
The problem of too many Z matrices being produced during multi-headed attention.
The requirement to install transformers via pip install -r requirements.txt.
In image generation, diffusion models address a major drawback of Generative Adversarial Networks (GANs) by simplifying the process into a kind of iterative small step situation. Which drawback is minimized by this iterative approach?
The need for large generator networks.
The necessity of text conditioning.
The difficulty and instability of training, including problems like mode collapse.
The requirement for Query, Key, and Value vectors.
In the Transformer encoder, which mechanism is employed to give the model a sense of the order of words in the input sequence, since the core attention mechanism lacks sequential awareness?
Multi-headed attention
Softmax calculation
Residual connections
Positional encoding
What happens to the eight separate SZS matrices produced by the multiple attention heads in the Transformer architecture before they are passed to the feed-forward layer?
They are averaged and then normalized.
They are multiplied by the Key matrix and then divided by eight.
They are concatenated and then multiplied by an additional weight matrix ( WO ).
They are passed directly to the decoder stack.
Who are the instructors listed for the short course, "ChatGPT Prompt Engineering for Developers," offered by DeepLearning.AI in partnership with OpenAI?
A) Isa Fulford and Yann LeCun
B) Andrew Ng and Geoffrey Hinton
C) Isa Fulford and Andrew Ng
D) Andrew Ng and Sam Altman
What are two key principles taught for writing effective prompts in the "ChatGPT Prompt Engineering for Developers" course?
Use complex language and utilize few-shot prompting
Write clear and specific instructions, and give the model time to think
Always use zero-shot prompting and define a persona
Minimize context and avoid using delimiters
Which of the following tasks is explicitly listed as a capability of Large Language Model (LLM) APIs demonstrated in the ChatGPT Prompt Engineering for Developers course?
Data Clustering
Transforming text
Running Principal Component Analysis (PCA)
Model Deployment via Flask
What is the primary architectural concept used by the Transformer model to account for the order of words in the input sequence?
Self-Attention Layer
Multi-Headed Attention
Positional Encoding
Feed-Forward Neural Network
In the context of machine learning models, what is the fundamental difference between a Generative model and a Discriminative model?
Discriminative models learn patterns in language; Generative models learn patterns in images.
Generative models use labeled data; Discriminative models use unlabeled data.
Generative models generate new data instances; Discriminative models classify or predict labels for data points.
Discriminative models require transformers; Generative models do not.
What technique is used to prevent an LLM model from getting confused between instructions and user input, thereby helping to mitigate hallucination?
Iterative Prompting
Few-shot Prompting
Delimiters
Context Expansion
What process describes how Diffusion models (like Stable Diffusion or DALL·E) generate an image from noise iteratively?
Training two competing networks (Generator and Discriminator) until equilibrium is reached.
Iteratively predicting the noise that was added to a noisy image and subtracting it to get closer to the original image.
Performing a linear schedule of noise addition using only Gaussian noise.
Using a single-step network to go directly from random noise to the original image.
Retrieval-Augmented Generation (RAG) is a technique designed to address several limitations of Foundation Models. Which limitation does RAG not specifically aim to address?
High computational cost of initial model pre-training.
Lack of depth in domain-specific knowledge.
Knowledge cutoffs leading to outdated information.
Hallucination and inability to cite sources.
What seminal paper introduced the Transformer model, which became the foundational architecture for modern Large Language Models?
Deep Residual Learning for Image Recognition
Attention Is All You Need
Generative Adversarial Networks
Playing Atari with Deep Reinforcement Learning
