Streaming responses in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main benefit of streaming responses in AI models?

Imagine you are chatting with an AI assistant that starts answering your question immediately, instead of waiting to finish the whole answer. What is the main benefit of this streaming response?

A It allows the AI to learn from your feedback instantly during the response.

B It improves the accuracy of the AI model's final answer.

C It reduces the waiting time by sending partial answers as soon as they are ready.

D It compresses the data to use less internet bandwidth.

💡 Hint

Think about how you feel when you get some information quickly instead of waiting for everything.

Streaming responses send parts of the answer as soon as they are generated, so you don't have to wait for the entire response to be ready. This reduces perceived waiting time.

❓ Predict Output

intermediate

What is the output of this streaming simulation code?

Consider this Python code simulating streaming output from an AI model. What will it print?

Prompt Engineering / GenAI

import time

def stream_response():
    parts = ['Hello', ', ', 'this ', 'is ', 'streamed ', 'text.']
    for part in parts:
        print(part, end='', flush=True)
        time.sleep(0.1)

stream_response()

A Hello, this is streamed text

B

Hello
, 
this 
is 
streamed 
text.

C .txet demaerts si siht ,olleH

D Hello, this is streamed text.

💡 Hint

Look at how print is used with end='' and flush=True.

The code prints each part without adding a new line, so the output is all parts joined without a trailing newline.

❓ Model Choice

advanced

Which model architecture is best suited for streaming text generation?

You want to build an AI that generates text word by word and sends each word immediately as it is created. Which model type is best for this streaming task?

A Transformer with full sequence attention computed before output

B Recurrent Neural Network (RNN) that processes input step-by-step

C Convolutional Neural Network (CNN) for image classification

D Feedforward Neural Network with no memory

💡 Hint

Think about which model can generate output one step at a time.

RNNs process input sequentially and can generate output token by token, making them suitable for streaming generation.

❓ Metrics

advanced

Which metric best measures the quality of streaming text generation?

When evaluating streaming text generation, which metric helps measure how well the generated text matches expected text over time?

A BLEU score comparing generated and reference text n-grams

B Mean Squared Error between predicted and actual token IDs

C Perplexity measuring how surprised the model is by the next token

D Accuracy of classifying text sentiment

💡 Hint

Think about comparing generated sentences to reference sentences.

BLEU score compares n-grams of generated text to reference text, useful for evaluating text generation quality.

🔧 Debug

expert

Why does this streaming code cause a delay before any output?

Look at this Python code meant to stream words with a delay between each. However, it waits fully before printing anything. Why?

Prompt Engineering / GenAI

import time

def stream_words(words):
    output = ''
    for w in words:
        output += w + ' '
        time.sleep(0.5)
    print(output)

stream_words(['This', 'is', 'streamed', 'text'])

A The print is outside the loop, so output prints only after all words are added.

B The time.sleep call blocks printing until the loop finishes.

C The output string is not flushed after each word, causing buffering.

D The function is missing a yield statement to stream words.

💡 Hint

Check where the print statement is placed relative to the loop.

The print happens after the loop ends, so all words are concatenated first, then printed at once, causing delay.