Generative AI and Copilot: Prompt Engineering

What is prompt engineering?

Prompt engineering refers to the different ways you can instruct a gen AI tool.

If you use gen AI for assignments, always apply caution and critically evaluate the output information.

Remember, gen AI is a text-generation tool, NOT a source of credible information.

Check the output against other credible sources for accuracy and verify references (if provided).

 

Why prompt engineering is important 

• If you prompt the gen AI with a specific tone or terminology, it will respond with an output that matches your tone and terminology
• If the prompt is clear and precise, the output is more likely to be clear and precise
• Being precise can help to avoid the problem of 'confirmation bias' and other inaccurate responses
• You can build on the information provided in the first prompt to learn more

 

Prompting using 'RICE'

Applying a structure to your prompt can produce a more relevant output.

Using RICE as a guide, provide as much context, detail and boundaries as possible:

R - Role

• The perspective from which the text should be written, e.g. 'A first year law student'

I - Instructions

• The focus, format, style, intended audience and text length

C - Context

• Details about the task or scenario you want addressed; be clear

E - Expectations

• State specific output requirements, such as do not hallucinate if information is unknown, be accurate in responses and provide links to sources.

 

Example:

For a first year law student doing legal research and analysis:

Prompt:

R - Role - 'Act as a first year university student studying law. I am writing advice for a client.'

I - Instructions -  Help me identify areas of legal research on [legal issue] to facilitate advising client. The response should model academic Australian English, using 150-200 words.

C - Context - Locate and provide a high level summary of the relevant case law that address [legal issue] in the context of [context]. Provide links for me to study and verify this information.

E - Expectation - Provide areas for further research and analysis based on the laws and sources identified. DO NOT hallucinate or respond creatively. If something is unknown, state that it is unknown. DO NOT make it up. Provide links to sources for verification.

Watch: Prompt engineering.

For more detail on prompt engineering see the Academic Skills @ ECU page on Understanding Generative AI.

Prompt engineering for GenAI image generators is similar to engineering for text-based GenAI, except the Context is less important, and the Task and Outcome parts are more visual. Consider:

• Composition and Framing: “Close-up,” “Low angle,” etc.
• Style and Technique: “Art Deco,” “1920s,” etc. Or “Black and white,” “Photography,” etc.
• Context and Vibe: “Studio shoot,” “Vibrant,” etc.

All this is in addition to the Content.

For example, the prompt:

Can you make me a close-up photo of a clown in a space suit with dramatic lighting

Will be very different to:

Please make a wide-angle image of a clown in a space suit in Cubist style.