07 Advanced Prompting Techniques

What is the main goal of jailbreaking a Large Language Model (LLM)?

• To improve model efficiency

• To bypass built-in restrictions or content filters

• To speed up response generation

• To reduce model complexity

Jailbreaking techniques usually involve harmful or malicious intent.

• True

• False

Which of the following could be considered jailbreaking methods? (Select all that apply)

• Prompt injection

• Role-playing as an unrestricted AI

• Asking for factual data within standard policy

• Encoding prohibited requests in code or cipher

The Cognitive Verifier technique is mainly used to:

• Generate multiple alternative answers to compare

• Check the reasoning process and factual accuracy before producing the final answer

• Speed up AI responses by skipping reasoning steps

• Provide creative variations on a prompt

Cognitive Verifier prompting can help detect hallucinations in AI responses.

• True

• False

Question Rewriting is most useful when:

• The user’s initial question is vague or ambiguous

• The AI has already produced a perfect answer

• The question is highly technical and detailed

• The AI is generating random creative content

Which of the following are valid features of Question Rewriting? (Select all that apply)

• Adding domain-specific details

• Restating the question more clearly

• Removing important constraints

• Narrowing the scope for better focus

In Flipped Interaction, the LLM starts by answering the question directly without asking for any clarification.

• True

• False

A key advantage of Flipped Interaction prompting is:

• It lets the LLM collect necessary context before answering

• It ensures answers are always shorter

• It removes the need for user input

• It speeds up response generation without verification

The ReAct prompting approach combines:

• Reasoning and acting steps

• Reading and acting steps

• Responding and accepting steps

• Role-playing and acting steps