RAG (Retrieval-Augmented Generation) QUIZ (MCQ QUESTIONS AND ANSWERS)

Using multiple retrieval steps to answer complex questions

Jumping between different models

Using multiple databases

Retrieving from multiple time periods

RAG updates model weights, fine-tuning doesn't

RAG uses external knowledge at inference, fine-tuning updates model parameters

RAG is faster than fine-tuning

Fine-tuning requires less data than RAG

Increased hallucination

Context window limitations and chunking trade-offs

Higher computational cost for generation

Slower embedding generation

Caching based on document titles

Compressing embeddings

Caching database queries

Storing responses for semantically similar queries

The speed of retrieval

How relevant retrieved contexts are to the user's question

The size of the context window

The number of documents retrieved

A method to compress RAG models

A technique that combines RAG with fine-tuning for domain-specific knowledge

A tool for database optimization

A benchmark for RAG evaluation

Sparse vectors are smaller in size

Sparse vectors are faster to process

Dense vectors encode meaning, sparse vectors encode word identity

Dense vectors contain more zeros

Combining multiple LLMs

Using both traditional text search and vector search

Mixing supervised and unsupervised learning

Using multiple databases simultaneously

Breaking a complex query into multiple simpler questions

Compressing queries to save tokens

Translating queries into different languages

Removing unnecessary words from queries

To remove corrupted data

To compress metadata

To generate new metadata

To narrow down retrieval based on document properties

RAG using multiple agents

RAG where an AI agent decides which tools to use and when

RAG for autonomous vehicles

RAG with human-in-the-loop

Answer faithfulness

Perplexity

BLEU score

F1 score

To ensure outputs follow safety and quality standards

To prevent system crashes

To limit database size

To increase processing speed

GLUE

RAGAS

SuperGLUE

HellaSwag

Lack of initial documents in the knowledge base

Slow initial model loading

Low temperature settings

Hardware initialization delays

By retraining the entire model

By updating the external knowledge base without retraining

By increasing model parameters

By using time-series analysis

It retrieves the k most recent documents

It retrieves k random documents

It retrieves the k most relevant documents based on similarity

It retrieves the k shortest documents

Retrieving documents multiple times to refine results

Using previous retrieval results to inform subsequent retrievals

Storing retrieval history

Repeating failed retrievals

How much of the ground truth is captured in retrieved context

How fast context can be retrieved

The size of retrieved context

The accuracy of the generator

To organize documents chronologically

To compress passages

To translate passages

To score and rerank retrieved passages for relevance

To compress the database

To convert text into numerical vector representations

To translate languages

To generate summaries

Encoder and Decoder

Trainer and Validator

Retriever and Generator

Parser and Compiler

Generate new data from scratch

Fetch relevant documents or information from a knowledge base

Train the generative model

Optimize the model's hyperparameters

Transformer-based large language model

Decision Tree

Linear Regression

K-Means Clustering

By incorporating external knowledge during generation

By reducing model size

By eliminating the need for training data

By increasing computational speed

Relational database

Graph database

Vector database

Document database

Faster inference speed

Access to updated and domain-specific information

Smaller model size

Lower computational cost

The model is fine-tuned

The retrieved context is added to the user's prompt

The database is updated

The model weights are adjusted

MySQL

MongoDB

Pinecone

Redis

Token length

Processing time

Hallucinations

Model parameters