Prompts for Creative Writing #2
Audience Persona and Few Shot Prompts
Having been inspired by this podcast to ensure that when prompting I focus more on the process rather than the output, I have been playing with several different types of prompts that can be used to support creative writing. I had some great feedback about my previous blog post, so I wanted to continue the series. Whilst the examples below are how I have used these for creative writing tasks, all the prompts can be adapted for any subject.
The two prompt techniques are called Audience Persona Prompts and Few Shot Prompting. I hope you find them helpful.
Audience Persona Prompt
Who is your audience? These prompts are a powerful way of ensuring your outputs align with the needs of your audience. In the classroom, this would most often be the pupils you are teaching. This can be particularly effective in differentiating the outputs based on a pupil’s needs.
Prompt Format:
Explain X to me.
Assume that I am person Y.
Now, you can iterate on this prompt. Note the difference when the audience’s persona is changed. The output for an EAL pupil is considerably different to the version above.
Perhaps, if using it for pupils who are EAL (English as an Additional Language), ask it for reasons as to why it chose certain words (See below). How do these words add to the mood? What would happen if you substituted the words? Dial up the temperature slightly and ask it for more powerful verbs.
Few Shot Prompting
This is a technique whereby a few examples are fed to the AI so that it is guided towards understanding what is expected in its output. However, the structure of the prompt is quite specific, with input and output being given.
Prompt Format:
Input: X
Output: Y
Initially, I was unsure how these prompts could best be deployed in the English classroom. There are numerous examples of mathematicians using it to improve the ability of AIs to solve word problems. So, I consulted Chat-GPT and it gave me some excellent examples. From there, I began to generate a few of my own.
Firstly, it can be used to generate ideas and support resources for setting descriptions. You will need to feed the AI a few examples of your own, and once it understands your train of thought, begin to provide specific locations/settings.
Secondly, it can be an effective way of generating ideas for writing poetry. Using the adjective/noun (e.g. Springtime blossoms; Midnight ocean) format in the input is very effective to getting the AI to generate vivid imagery.
Thirdly, it could be used to generate ideas for a story. You could prompt the LLM to help generate ideas for any part of a potential story. It is a common occurrence that a child says they do not know how to start or end a story. You could work with the pupil to think about the parts of their story they do know. Do they have ideas for characters? Where is the story taking place? What is the climax of the story? From there, you could structure few shot examples to help with idea generation.
Finally (and my favourite), use few shot examples to generate the mood/atmosphere based on lines from books or poems. This would serve as an excellent discussion point. Do they agree/disagree? Is there enough detail?
With these examples, pupils can test their own sentences and see how small changes to vocabulary can drastically alter the mood/atmosphere. The below example changes the verb ‘charged’ to ‘ambled’ and pupils can quickly see the resultant impact of a single alteration of vocabulary.
Final Thought
I really do not think you need to be aware of specific names for prompts. The important part of working with AI is simply engaging in the process of prompting and seeing the outcomes. However, I have found that using specific techniques and then adapting as I prompt, has been a helpful way to structure my thinking. Because let’s be honest, we have all sat in front of an AI and been clueless about what to input. So, we end up with another poem.