Prompt essentials
Getting the best possible content from GenAI tools means being aware of the what, the why and the how of prompt engineering. As GenAI models enter everyday life, prompting skills will become more in-demand. For students, being able to work effectively with GenAI directly impacts successful study and learning experiences.
Becoming a prompt expert means having a clear working idea of what prompts, outputs and prompt engineering are. These concepts are explored in more depth in this guide. Get started with the following three key definitions.
Prompt
Prompts in GenAI are the questions we ask to get specific content or answers from the computer.
Output
Outputs in GenAI are the content or answers the computer gives based on our questions.
Prompt engineering
Prompt engineering for GenAI is designing the right questions to get the best content or answers from an artificial intelligence.
Elements of prompting
What are prompts
Prompts are commands you use for generative AI tools. Prompts are typically word based. However, you can also provide visual or audio prompts in written instruction in some GenAI tools. They can be questions, files, images or other data that the tool responds to when producing output. Prompts or commands are crucial to working with technology. They’re the instructions you use to interact with the program or system. Think of how you ask Google or Library Search to find information you need. Or the way you request Siri, Alexa or other virtual assistants to do something for you. They have a huge influence on the quality of the material you generate.
Be careful with the prompts you provide:
- Overgeneralised prompts can produce output that is too broad to be useful.
- Overly specific prompts can produce output that is too specific and misses the big picture.
- Ambiguous or unclear prompts can produce vague and unusable output.
- Malicious by design prompts produce deliberately inaccurate or biased output that in turn feeds back into the system (poisoned data).
Well crafted prompts are the grounding for good quality content to be generated. However, just like when you are drafting something it takes a few versions. Usually, a series of prompts are used in conversation with the GenAI tool to get a good result.
Specificity is key! When crafting prompts, remember to provide context and constraints, such as asking for a finite set of results.
What are outputs
Outputs are the results or content GenAI tools produce, in response to the input elements or prompts you provide. Outputs can be text, image, or audio based (or a combination). Outputs are not fixed or final and can be edited, modified, or improved. The quality of the prompt input used with the GenAI tool will directly impact on the quality and effectiveness of an output.
Outputs can be used for various purposes, such as entertainment, education, research, or business.
What is prompt engineering
Prompt engineering stands as a powerful tool in equipping us to apply the critical skills necessary to navigate and thrive in this rapidly changing world.
Prompt engineering is about asking the right clear and concise structured questions in order to obtain a clear focused relevant response.
What makes a good prompt is the art of prompt engineering. Prompt engineering is the practice of building and refining prompts to ensure quality output is produced by the GenAI tool. Crafting clear and effective instructions or questions for GenAI tools helps them to produce content that matches your expectations.
Taking some time to consider the various elements of the content you would like the GenAI tool to produce for you will help you to develop a more effective prompt. Crafting an effective prompt is similar to following a recipe to cook a meal. Using quality ingredients and the right cooking techniques will help you to produce a delicious meal.
Just like cooking can be challenging and finding the best recipe takes experimentation and time, engineering a good prompt is the same.
Understanding the strengths and limitations of your chosen AI tool is key to formulating a good prompt. Although, as AI becomes more advanced, there will be less need to be meticulous about creating the perfect prompt and new capabilities may instead be required.
Prompt Engineering
Good practice to GenAI prompting
Specificity and revision are key to prompt engineering. Click on the plus (+) icons in the below interactive infographic for guidance on how to prompt GenAI to get what you need.
Activity overview
This interactive image hotspot provides a process for good practice when it comes to prompt engineering. Hotspots are displayed as plus (+) icons that can be clicked, to present the information.
Hotspot 1
Define a Role
Set the right tone for the GenAI tool. Instruct the GenAI tool to play a specific role such as a coach advisor tutor or creator. This ensures the information is communicated in a particular style, the right level of detail, and using the appropriate language.
GenAI tools can take on various personas, jumping between different professions and different roles.
Hotspot 2
Give a Task
State the task you want completed by the GenAI tool. The task is a summary of what you need the GenAI tool to create.
Provide the GenAI with tool specific details of what you need - specific focus areas. Clearly communicating the goal of the task helps the GenAI tool to produce a more targeted output.
Hotspot 3
Establish Context
Give context to the GenAI tool. Context is background information that helps the GenAI tool better understand how to follow instructions, complete tasks, and avoid making assumptions.
Context can include your audience, geographic or demographic impacts, ethical or cultural sensitivities and outputs format, style, genre or mood.
Hotspot 4
Try Prompt
Once you have constructed a prompt you then need to run it in a generative AI tool.
Hotspot 5
Evaluate Outcome
The first time that you prompt a generative AI tool, it is unlikely to produce your desired outcome. After you have run your prompt, you need to evaluate what the GenAI tool produced based on the prompt.
Hotspot 6
Refine
If the GenAI tool did not produce to the desired outcome, it means something in your prompt was not clear. You can refine your prompt until you get the outcome you are looking for.
Hotspot 7
Outcome Achieved
The GenAI tool produces the outcome that you were after based on the prompt you gave it.
Different content needs different prompts
Creating various content types (text, visual, audio) often requires different GenAI tools. Each tool demands unique prompt strategies to achieve effective and desired results.
Let's break down some of the considerations for each.
Clarity: Your prompt should be clear and specific. Vague prompts can lead to vague outputs.
Leading: You can set the tone, style, or context in your prompt. For example, 'write a horror story about a haunted piano' versus 'Write a romantic story about a piano in a Parisian cafe.'
Constraints: You can limit or guide the model's response by adding constraints, such as word limits or specific structures.
Descriptive: Since you are trying to visualise a concept, your description needs to be detailed. For exmaple, 'A serene sunset over a mountainous landscape with a river flowing in the foreground.'
Input Images: Some models allow you to input an image as a starting point, and then describe modifications or elements you want to add or change.
Iteration: You might not get the perfect image on the first try. Refining your prompts or giving feedback on initial outputs can help in achieving the desired result.
Descriptive Soundscapes: Describe the audio environment you want. Fore example 'A busy city street with distant sounds of a jazz band, honking cars, and nearby footsteps.'
Musical Details: If generating music, you might specify instruments, tempo, mood, or genre. For example, 'A calm piano melody with a slow tempo, accompanied by a soft cello in the background.'
Duration: You may need to specify the length of the audio clip.
Feedback Loops: For nuanced compositions, iterative feedback might be required.
For some further prompt engineering strategies and tactics for large language models, visit the prompt engineering guide by Open AI.
Acknowledgement
Information on this guide is adapted from Deakin University using a CC BY-NC 4.0 license.