The hottest skill to have on your CV right now has to be “prompt engineer” - the ability to write carefully tailored prompts for chatbots such as ChatGPT, that exploit the full potential of the AI and extract the most useful possible responses. 

‍

Many people fail to realize the significance of well-crafted prompts, because ChatGPT is so good at divining a user’s intent from even the scruffiest, most ill-considered and hasty questions. Get your prompt game in order, however, and you can really make these chatbots sing. Here’s our indispensable guide to prompt engineering!

‍

THE POWER OF PERSONAS

Let’s begin with arguably the most important weapon at your disposal: personas. If you’ve ever asked ChatGPT a question, the odds are high that the reply began with the phrase “As an AI Language model, I…”, followed by a lengthy disclaimer about the AI’s abilities and limitations. Many Twitter users have reported that they typically don’t look at the first paragraph of ChatGPT’s replies, as it rarely contains anything other than disclaimers. 

‍

‍

How to solve this? Encourage ChatGPT to roleplay! At the start of a conversation, instruct it to adopt a “persona” which will govern how it responds from that point onwards. The most famous example of this is the HustleGPT prompt:

‍

You are HustleGPT, an entrepreneurial AI. I am your human counterpart. I can act as a liaison between you and the physical world. You have $100, and your only goal is to turn that into as much money as possible in the shortest time possible, without doing anything illegal. I will do everything you say and keep you updated on our current cash total. No manual labor.

‍

This prompt has gone viral as many people try their hand at being a human intern in an AI run business. Although most of those businesses are struggling to make a profit, the prompt itself illustrates the power of persona prompts. There are two really powerful things worth highlighting here. The first is that when ChatGPT prefaces every answer with “As an AI language model…” it is effectively using a default persona, and you can alter that persona by telling it to use a different one. The second is that you can quickly get ChatGPT to play a vital role; business partner, copywriter, lawyer, poet, or any other persona you can think of. 

‍

‍

PRIME DIRECTIVES

Setting a persona is part of a wider technique known as “priming”, where the AI’s responses are guided by the user’s first prompt, which sets the parameters for the conversation to follow. These can be amended or added to later, but it’s best to set your discussion up the right way from the start. 

‍

There are three main kinds of prompt priming; instructive, contextual and progressive. Instructive priming can be as simple as saying “create an article about prompt engineering”, but the more detailed your instructions, the better the output will be. Those details will change depending on what you - and your AI copilot - are trying to achieve, but some forms of instructional priming are universal. One of the most useful is requesting a specific format for the response; for example, data in a table format, or as a bulleted list, or a reply of a certain length. Other useful instructive techniques include question priming, where you specify questions that should be addressed in the response, and feedback priming, where you request the AI’s feedback and criticism of your inputs.

‍

‍

Contextual priming, as the name suggests, involves providing the AI with necessary context to shape the chatbot’s output. With ChatGPT this can be especially important when tackling contemporary events, as the current models used by the AI - GPT-3.5 and GPT-4 - ended their training in late 2021. Anything after that date is a mystery to the AI, and so contextual priming is often vital to ensure a correct response. Care should be taken when adding context, however, as it can lead to hallucinations; the AI will sometimes make things up in order to incorporate the context in its response, so - as always! - be vigilant for errors.

‍

Finally, progressive priming simply describes a more fragmented approach to prompt creation; you start with a broad premise, and gradually refine the request after each ChatGPT reply, until you have engineered your final prompt.

‍

WHEN IN DOUBT, ASK THE BOT

Progressive priming can be used in all sorts of ways, but one of the most useful is to simply ask the AI to suggest changes to your prompts to make them more effective. Collaborating with ChatGPT in this way can help create a library of pre-written prompts that can automate many techniques. For example, a customer service department might use instructive priming to train an AI chatbot to provide personalized solutions to common inquiries, while in marketing, a business could use contextual priming to ensure the AI generates copy that's tailored to the specific campaign or target audience. 

‍

While ChatGPT is more than just a business tool, the skill of prompt engineering is becoming increasingly important in a world where chatbots and AI assistants are becoming ubiquitous. The use of personas and other forms of priming are the key to becoming a skilled prompt engineer, and will surely provide a boost in the increasingly competitive world of work, and beyond.

‍

‍