By Phil Spalding

cambolc.co.uk,  LinkedIn, BlueSky, Facebook

Meet the AI

Words of the moment are Artificial Intelligence. Most people have used AI if they have an Alexa or virtual assistant. A Large Language Model (LLM) is used to underpin the interaction. We all know what happens when we come up with a request that Alexa cannot process. Alexa responds by telling you, “Sorry, I don’t know that”. Other times the question or prompt may provide information that you did not request or you know is patently false. The Human in the Loop (HITL), yourself, will then try to refine the prompt to obtain the information that is useful. Alexa does not yet have the Australian Siri satirical experience for poorly crafted prompts. Different AIs have different ways of working. However, no matter which AI you use, the magic of AI is all in the prompt.

Building Powerful Prompts

(Accompanying Slide Show)

So how do you write a good prompt? The best practice habits come from understanding that:

• 01 The prompt is the heart of AI Communication
• 02 Think of it like giving directions: the more specific, the easier to reach the destination
• 03 Provide clear instructions to get what yo want
• 04 The more details you provide in your prompt, the better the output.

A plan or scaffold on which to build our enquiry is more likely to lead to the right destination. It may take several refinements or iterations to generate the right information in the opinion of the HITL. Best practice is to build your prompt using 5 P.A.R.T.S.

• Persona: Define your role
• Aim: State your objective clearly
• Recipients: Specify your target audience
• Theme: Describe the desired style and tone.
• Structure: Specify the output format

Whichever AI you choose to use (e.g. ChatGPT or Gemini) be aware you may not get what you need from the AI. Practice makes perfect.

Responsible Use of AI

Further habits that should underpin best practice in AI are:

• Review AI Outputs: Verify accuracy and personalise content.
• Thoughful Use: Use your judgement and avoid harm.
• Disclose AI Use: Be Transparent about using AI.
• Privacy and Security: Protect sensitive information

Next Steps

I suggest experimenting with using prompts. Find out how your chosen AI responds. What provokes hallucinations? This is the process where a particular word or phrase produces an errant response. Some good examples can be found here.

Think about making a series of common templates before you launch into using the AI based on the 5 P.A.R.T.S. principles ( see above for slideshow). Make them relevant to you particular expertise and style.

The last steps before sharing your carefully crafted prompt results, is to do no harm. Make sure that the HILT has viewed, understood the information and preserved security of data. Legal jeopardy could attach here.

Safe prompting!