NEW
Font size
WorksheetsAI Concepts Quiz: Questions 1-10
Total questions: 25
Worksheet time: 13mins
Transformers introduced which core concept enabling models to focus on relevant tokens?
Pooling
Self-Attention
Activation
Which transformer block is responsible for generating outputs token-by-token?
Encoder
Decoder
Feed-Forward
Which model is encoder-only?
T5
BERT
GPT
Large Language Models are primarily trained on:
Transaction logs
Massive text datasets
Image datasets
In Generative AI, inference refers to:
Model initialization
Generating responses to new inputs
Loss minimization
Which term best describes Generative AI models creating new content?
Creative generation
Reconstructive
Predictive
The main goal of RAG is to solve:
High training costs
Hallucination and outdated knowledge
Tokenization errors
In RAG, which component retrieves supporting documents?
Generator
Tokenizer
Retriever
Embeddings convert text into:
JSON
HTML
Numerical vectors
Similar vectors in an embedding space indicate:
Incorrect tokenization
Semantically related meanings
Randomly distributed tokens
Vector databases are mainly used for:
Semantic similarity search
Storing raw PDFs
Grammar correction
In training, the objective is to:
Remove embeddings
Minimize loss
Increase number of tokens
Backpropagation adjusts:
Model weights
Dataset labels
Token IDs
Which optimizer is adaptive and widely used?
Newton
SGD
Adam
Overfitting means:
Model generalizes well
Model memorizes training data
Model does not learn enough
Function calling allows AI to:
Improve batch size
Execute real-world actions via APIs
Increase token length
First step in function calling flow:
Model formats response
User sends a query
API executes
Multi-agent systems rely on:
Collaboration among multiple agents
Single model decision
Static prompts
Which is an example of agent roles?
Deployer agent
Retriever & Summarizer agents
Billing agent
MAS communication commonly uses:
Binary encodings
Structured messages / JSON
Images
Positional encoding in transformers helps:
Train without embeddings
Add token order information
Increase sequence length
Granite LLM series focuses mainly on:
Gaming applications
Enterprise-grade, trusted AI
Crypto analytics
In RAG architecture, the knowledge base stores:
Temporary token scores
Real factual documents
Random synthetic prompts
Chunking documents before embedding helps:
Increase token repetition
Maintain context and avoid truncation errors
Reduce inference cost fully
Evaluation metrics like BLEU and ROUGE apply to:
Image recognition
Password encryption
Text-based models
