3. Applications of Foundation Models Flashcards
(25 cards)
What is Retrieval Augmented Generation (RAG)?
A) A technique for generating new data
B) A method of combining retrieved information with model generation
C) A type of model architecture
D) A data compression algorithm
B
RAG is a method of combining retrieved information with model generation.
Which AWS service is suitable for storing embeddings in a vector database?
A) Amazon S3
B) Amazon RDS
C) Amazon OpenSearch Service
D) Amazon EC2
C
Amazon OpenSearch Service is mentioned as a service for storing embeddings in vector databases.
What is the primary purpose of adjusting the temperature parameter in inference?
A) To control the physical temperature of the server
B) To adjust the creativity or randomness of the model’s output
C) To increase the model’s processing speed
D) To reduce energy consumption
B
The temperature parameter affects the creativity or randomness of the model’s output.
What is a chain-of-thought prompt?
A) A physical chain used in AI hardware
B) A prompt that encourages the model to show its reasoning process
C) A method of linking multiple AI models
D) A technique for encrypting prompts
B
Chain-of-thought is a prompt engineering technique that encourages the model to show its reasoning process.
Which of the following is NOT a typical method for fine-tuning a foundation model?
A) Instruction tuning
B) Transfer learning
C) Physical tuning
D) Continuous pre-training
C
Physical tuning is not a method for fine-tuning foundation models.
What is the ROUGE metric used for in evaluating foundation models?
A) Measuring the redness of the model’s output
B) Evaluating the quality of generated summaries
C) Calculating the model’s energy efficiency
D) Determining the model’s processing speed
B
ROUGE (Recall-Oriented Understudy for Gisting Evaluation) is used for evaluating the quality of generated summaries.
What is the primary purpose of using Agents for Amazon Bedrock?
A) To hire human agents for AI tasks
B) To handle multi-step tasks in AI applications
C) To physically maintain AI hardware
D) To reduce the cost of AI services
B
Agents for Amazon Bedrock are used to handle multi-step tasks in AI applications.
Which of the following is a key consideration when selecting a pre-trained model?
A) The model’s popularity on social media
B) The physical size of the server hosting the model
C) The model’s input/output length capabilities
D) The color scheme of the model’s documentation
C
The model’s input/output length capabilities are a key consideration when selecting a pre-trained model.
What is prompt hijacking in the context of prompt engineering?
A) A method of optimizing prompts
B) A technique for stealing prompts from competitors
C) An attack where the model is tricked into ignoring the intended prompt
D) A way to speed up prompt processing
C
Prompt hijacking is a risk where the model is tricked into ignoring the intended prompt.
What is the primary goal of instruction tuning in foundation models?
A) To teach the model to follow specific instructions
B) To reduce the model’s size
C) To increase the model’s processing speed
D) To change the model’s programming language
A
Instruction tuning aims to teach the model to follow specific instructions.
What is BERTScore used for in evaluating foundation models?
A) Measuring the model’s energy efficiency
B) Evaluating the quality of generated text
C) Calculating the model’s processing speed
D) Determining the model’s market value
B
BERTScore is used for evaluating the quality of generated text.
What is a key benefit of using in-context learning for foundation model customization?
A) It requires no additional training data
B) It always produces perfect results
C) It reduces the model’s size
D) It eliminates the need for prompts
A
In-context learning allows for model customization without additional training data.
What is a potential risk of using zero-shot learning in prompt engineering?
A) The model may perform poorly on tasks it wasn’t explicitly trained for
B) The model will refuse to generate any output
C) The model will only work with numerical data
D) The model will consume excessive energy
A
Zero-shot learning may result in poor performance on tasks the model wasn’t explicitly trained for.
What is the primary purpose of reinforcement learning from human feedback (RLHF) in foundation model training?
A) To reduce the model’s energy consumption
B) To improve the model’s performance based on human evaluations
C) To increase the model’s size
D) To translate the model into different languages
B
RLHF is used to improve the model’s performance based on human evaluations.
Which of the following is NOT a typical consideration when preparing data for fine-tuning a foundation model?
A) Data curation
B) Data size
C) Data labeling
D) Data color coding
D
Data color coding is not a typical consideration.
What is prompt templating in the context of prompt engineering?
A) A method of physically printing prompts
B) A technique for creating reusable prompt structures
C) A way to encrypt prompts
D) A process of translating prompts into different languages
B
Prompt templating is a technique for creating reusable prompt structures.
What is the primary advantage of using few-shot learning in prompt engineering?
A) It requires no examples in the prompt
B) It allows the model to learn from a small number of examples
C) It always produces perfect results
D) It reduces the model’s energy consumption
B
Few-shot learning allows the model to learn from a small number of examples.
Which AWS service is suitable for storing embeddings in a relational database?
A) Amazon DynamoDB
B) Amazon S3
C) Amazon Aurora
D) Amazon EC2
C
Amazon Aurora is mentioned as a service for storing embeddings in databases.
What is a key consideration when evaluating whether a foundation model effectively meets business objectives?
A) The model’s popularity on social media
B) The physical size of the server hosting the model
C) The model’s impact on user engagement
D) The color scheme of the model’s user interface
C
The model’s impact on user engagement is a key consideration when evaluating business effectiveness.
What is the primary purpose of negative prompts in prompt engineering?
A) To make the model generate negative emotions
B) To tell the model what to avoid in its output
C) To reduce the model’s energy consumption
D) To decrease the model’s processing speed
B
Negative prompts are used to tell the model what to avoid in its output.
What is continuous pre-training in the context of foundation models?
A) A method of constantly retraining the model on new data
B) A technique for training models 24/7
C) A way to train models using continuous mathematics
D) A process of training models on a continuous physical surface
A
Continuous pre-training involves constantly retraining the model on new data.
What is prompt poisoning in the context of prompt engineering risks?
A) A method of optimizing prompts
B) A technique for improving prompt quality
C) An attack where malicious content is inserted into training data or prompts
D) A way to speed up prompt processing
C
Prompt poisoning is an attack where malicious content is inserted into training data or prompts.
What is the BLEU score used for in evaluating foundation models?
A) Measuring the model’s energy efficiency
B) Evaluating the quality of machine translations
C) Calculating the model’s processing speed
D) Determining the model’s market value
B
BLEU (Bilingual Evaluation Understudy) is used for evaluating the quality of machine translations.
What is a key benefit of using transfer learning for foundation model customization?
A) It requires no additional training
B) It allows the model to leverage knowledge from one domain to another
C) It always produces perfect results
D) It reduces the model’s size to zero
B
Transfer learning allows the model to leverage knowledge from one domain to another.
