CAIC CAIC Natural Language Processing & AI Applications 2 — Questions and Answers
Question 1: Which technique allows an LLM to reference external documents at inference time without retraining?
- Fine-tuning
- Retrieval-Augmented Generation (RAG) (Correct answer)
- Transfer learning
- Quantization
Correct answer: Retrieval-Augmented Generation (RAG)
RAG combines a retrieval system with a generative model, allowing the LLM to fetch relevant documents and incorporate them into its response without retraining.
Question 2: What is 'hallucination' in the context of large language models?
- Model overfitting to training data
- Model generating confident but factually incorrect outputs (Correct answer)
- Model refusing to answer sensitive questions
- Model outputting duplicate tokens
Correct answer: Model generating confident but factually incorrect outputs
LLM hallucination refers to the model producing plausible-sounding but factually incorrect or fabricated information with apparent confidence.
Question 3: In a named entity recognition (NER) task, which of the following would typically be tagged as an entity?
- Conjunctions and prepositions
- Person names, organizations, and locations (Correct answer)
- Verb tenses and grammatical structures
- Stop words like 'the' and 'a'
Correct answer: Person names, organizations, and locations
NER identifies and classifies named entities in text into categories such as person names, organizations, locations, dates, and monetary values.
Question 4: Which approach is used to reduce a pre-trained LLM's size while preserving most of its performance?
- Data augmentation
- Model quantization (Correct answer)
- Ensemble stacking
- Oversampling
Correct answer: Model quantization
Quantization reduces model size by lowering the numerical precision of weights (e.g., from 32-bit to 8-bit), enabling deployment on resource-constrained hardware.
Question 5: What is a vector embedding in NLP applications?
- A compressed binary representation of images
- A dense numerical representation of text capturing semantic meaning (Correct answer)
- A rule-based regex pattern for text matching
- A database index for full-text search
Correct answer: A dense numerical representation of text capturing semantic meaning
Vector embeddings map words, sentences, or documents into high-dimensional numerical vectors where semantically similar content has closer proximity.
Question 6: Which task involves training an AI model to answer questions based on a provided passage of text?
- Text summarization
- Machine translation
- Extractive question answering (Correct answer)
- Part-of-speech tagging
Correct answer: Extractive question answering
Extractive question answering locates and extracts spans of text from a provided context passage that directly answer a given question.
CAIC CAIC Natural Language Processing & AI Applications 2
Which technique allows an LLM to reference external documents at inference time without retraining?