Combining retrieved context with LLM in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

How does retrieved context improve LLM responses?

Imagine you ask a large language model (LLM) a question. How does adding retrieved context from a database help the LLM give better answers?

A It reduces the LLM's vocabulary size to speed up response time.

B It provides extra relevant information so the LLM can generate more accurate and specific responses.

C It replaces the LLM's internal knowledge completely with the retrieved data.

D It forces the LLM to ignore the question and only use the retrieved context.

💡 Hint

Think about how having more facts related to your question helps you answer better.

Retrieved context adds relevant facts or documents that the LLM can use to generate more accurate and detailed answers. It supplements the LLM's internal knowledge without replacing it.

❓ Predict Output

intermediate

Output of combining retrieved context with LLM prompt

Given the following Python code snippet that combines retrieved context with a prompt for an LLM, what is the printed output?

Prompt Engineering / GenAI

retrieved_context = "The Eiffel Tower is in Paris."
prompt = "Where is the Eiffel Tower located?"
combined_input = f"Context: {retrieved_context}\nQuestion: {prompt}"
print(combined_input)

A

Context: The Eiffel Tower is in Paris.
Question: Where is the Eiffel Tower located?

B Context: The Eiffel Tower is in Paris. Question: Where is the Eiffel Tower located?

C

Context: retrieved_context
Question: prompt

D The Eiffel Tower is in Paris. Where is the Eiffel Tower located?

💡 Hint

Look carefully at how the f-string formats the variables with newlines.

The f-string inserts the retrieved context and prompt with a newline between them, so the output shows 'Context: ...' on one line and 'Question: ...' on the next.

❓ Model Choice

advanced

Best model architecture for combining retrieved context with LLM

Which model architecture is best suited to effectively combine retrieved context with a large language model for question answering?

A A recurrent neural network that processes only the retrieved context without the question.

B A simple feedforward neural network that ignores context and only processes the question.

C A convolutional neural network trained on images unrelated to text.

D A retrieval-augmented transformer that encodes context and question jointly before decoding.

💡 Hint

Think about architectures designed to handle both context and question together.

Retrieval-augmented transformers are designed to jointly encode retrieved context and the question, enabling the model to generate answers based on both inputs effectively.

❓ Hyperparameter

advanced

Choosing the number of retrieved documents for context

When combining retrieved context with an LLM, what is a key consideration when choosing how many documents to retrieve?

A Retrieving documents unrelated to the question improves diversity and accuracy.

B Retrieving only one document always guarantees the best answer.

C Retrieving too many documents can overwhelm the model and reduce answer quality due to noise.

D Retrieving zero documents is best because the LLM knows everything already.

💡 Hint

More context is not always better; think about information overload.

Retrieving too many documents can introduce irrelevant information, confusing the model and lowering answer quality. A balanced number of relevant documents is ideal.

❓ Metrics

expert

Evaluating combined retrieval and LLM system performance

You have a system that retrieves documents and then uses an LLM to answer questions. Which metric best measures how well the combined system answers questions accurately?

A Exact Match (EM) score comparing generated answers to ground truth answers.

B Mean Squared Error (MSE) between retrieved document vectors.

C BLEU score comparing retrieved documents to questions.

D Training loss of the LLM on unrelated text data.

💡 Hint

Think about metrics that compare generated answers to correct answers.

Exact Match (EM) measures the percentage of generated answers that exactly match the correct answers, making it a good metric for question answering accuracy.