CAIC Natural Language Processing & AI Applications Flashcards

Which technique allows an LLM to reference external documents at inference time without retraining?

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.

What is 'hallucination' in the context of large language models?

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.

In a named entity recognition (NER) task, which of the following would typically be tagged as an entity?

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.

Which approach is used to reduce a pre-trained LLM's size while preserving most of its performance?

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.

What is a vector embedding in NLP applications?

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.

Which task involves training an AI model to answer questions based on a provided passage of text?

Answer: Extractive question answering

Extractive question answering locates and extracts spans of text from a provided context passage that directly answer a given question.