Why AI safety prevents misuse in Prompt Engineering / GenAI - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why is AI safety important to prevent misuse?

Imagine you have a smart assistant that can do many tasks. Why do we need AI safety rules to stop it from being used in harmful ways?

A Because AI safety makes the AI ignore user commands to prevent misuse.

B Because AI safety ensures the AI always follows ethical guidelines and avoids harmful actions.

C Because AI safety allows the AI to learn without any limits or controls.

D Because AI safety makes the AI faster and more powerful for all tasks.

💡 Hint

Think about how rules help keep things safe in real life, like traffic laws.

AI safety focuses on making sure AI systems act in ways that are ethical and do not cause harm. This helps prevent misuse by bad actors or accidental damage.

❓ Model Choice

intermediate

Choosing a model to reduce misuse risk

You want to build an AI chatbot that avoids giving harmful advice. Which model choice helps reduce misuse risk?

A A model that ignores user inputs and gives random answers.

B A model trained only on large internet data without filtering harmful content.

C A model trained with reinforcement learning from human feedback to follow safety guidelines.

D A model trained to maximize engagement regardless of content safety.

💡 Hint

Think about how human feedback can teach AI to be safer.

Training with human feedback helps the AI learn to avoid harmful or unsafe responses, reducing misuse risk.

❓ Metrics

advanced

Measuring AI safety effectiveness

You want to check if your AI system is safe and not misused. Which metric best measures this?

A The percentage of AI responses flagged as harmful or unsafe by users.

B The AI model's training loss value after each epoch.

C The total number of AI responses generated per second.

D The AI system's CPU usage during inference.

💡 Hint

Think about how to detect harmful outputs from the AI.

Tracking how often AI outputs are flagged as harmful helps measure how well safety measures prevent misuse.

🔧 Debug

advanced

Debugging misuse in AI output filtering

Given this code snippet that filters harmful AI outputs, which option explains why harmful content still appears?

def filter_output(text):
    harmful_words = ['hack', 'attack', 'steal']
    for word in harmful_words:
        if word in text:
            return 'Content blocked due to safety.'
    return text

output = filter_output(ai_response)

A The filter only checks the first word of the text for harmful content.

B The filter blocks all outputs containing harmful words correctly.

C The filter returns the original text even if harmful words are found.

D The filter misses harmful words if they appear with different letter cases like 'Hack' or 'ATTACK'.

💡 Hint

Think about how text matching works with uppercase and lowercase letters.

The filter checks for exact matches with lowercase words, so it misses harmful words with uppercase letters, allowing misuse.

❓ Hyperparameter

expert

Adjusting hyperparameters to improve AI safety

You train a language model and want to reduce the chance it generates harmful content. Which hyperparameter adjustment helps most?

A Lowering the temperature value during text generation to make outputs more focused and less random.

B Increasing the learning rate to speed up training and produce more diverse outputs.

C Increasing the batch size to use more data at once without changing output randomness.

D Setting the dropout rate to zero to prevent any neuron from being ignored.

💡 Hint

Think about how randomness affects the safety of generated text.

Lowering temperature reduces randomness, making the model less likely to produce unexpected or harmful outputs.