Content filtering in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main purpose of content filtering in AI?

Content filtering is used in AI systems to manage what kind of content is shown or blocked. What is the main goal of content filtering?

A To prevent harmful or inappropriate content from being shown

B To improve the speed of AI model training

C To increase the size of the training dataset

D To reduce the number of model parameters

💡 Hint

Think about why some AI systems block certain words or images.

Content filtering helps AI systems avoid showing harmful, offensive, or inappropriate content to users, ensuring safer and more responsible AI use.

❓ Predict Output

intermediate

What is the output of this content filtering code snippet?

Given the following Python code that filters out banned words from a list of messages, what is the output?

Prompt Engineering / GenAI

banned_words = {'bad', 'ugly'}
messages = ['This is good', 'This is bad', 'Ugly truth']
filtered = [msg for msg in messages if not any(word.lower() in banned_words for word in msg.split())]
print(filtered)

A ['This is bad', 'Ugly truth']

B ['This is good']

C ['This is good', 'Ugly truth']

D ['This is good', 'This is bad']

💡 Hint

Check which messages contain words in the banned_words set.

The code removes messages containing any banned word. 'This is bad' contains 'bad' and 'Ugly truth' contains 'Ugly' (case insensitive), so only 'This is good' remains.

❓ Model Choice

advanced

Which model type is best suited for real-time content filtering in chat applications?

You want to filter harmful content in a chat app instantly as users type. Which AI model type is best for this task?

A Lightweight rule-based model with keyword matching

B Large batch-trained transformer model with high latency

C Heavy deep learning model requiring GPU inference

D Offline unsupervised clustering model

💡 Hint

Consider speed and resource use for real-time filtering.

Lightweight rule-based models with keyword matching are fast and efficient for real-time filtering, unlike large or heavy models that cause delays.

❓ Hyperparameter

advanced

Which hyperparameter adjustment helps reduce false positives in a content filtering model?

You have a content filtering model that blocks too many safe messages (false positives). Which hyperparameter change can help reduce this?

A Lower the classification threshold

B Increase the learning rate

C Increase the classification threshold

D Decrease the batch size

💡 Hint

Think about how threshold affects model sensitivity.

Increasing the classification threshold makes the model more strict before blocking content, reducing false positives but possibly increasing false negatives.

🔧 Debug

expert

Why does this content filtering code fail to block the word 'Bad'?

Examine the code below. It should block messages containing the word 'bad' regardless of case, but it does not block 'Bad'. Why?

Prompt Engineering / GenAI

banned_words = {'bad'}
message = 'This is Bad'
if any(word.lower() in banned_words for word in message.split()):
    print('Blocked')
else:
    print('Allowed')

A The any() function is used incorrectly

B The banned_words set should include 'Bad' with uppercase B

C The split() method does not split the message correctly

D The code does not convert message words to lowercase before checking

💡 Hint

Check how case sensitivity affects membership tests.

The code checks words as-is against banned_words which only has lowercase 'bad'. 'Bad' with uppercase B does not match, so it is not blocked.