Generative arificial intelligence (AI) burst onto the scene almost 12 months ago; until then AI was for scientists and geeks. Today it has revolutionised the business and professional landscapes.
But as yet, attempts to fully explore the technology’s potential to reshape business functions have barely scratched the surface.
Generative AI systems fall under the broad category of machine learning, and here’s how one such system—ChatGPT—describes what it can do:
“Ready to take your creativity to the next level? Look no further than generative AI! This nifty form of machine learning allows computers to generate all sorts of new and exciting content, from music and art to entire virtual worlds. And it’s not just for fun—generative AI has plenty of practical uses too, like creating new product designs and optimizing business processes. So why wait? Unleash the power of generative AI and see what amazing creations you can come up with”
That’s why ChatGPT—the GPT stands for generative pretrained transformer—is receiving so much attention right now. It’s a free chatbot that can generate an answer to almost any question it’s asked. Developed by OpenAI, and released for testing to the general public in November 2022, it’s already considered the best AI chatbot ever.
And it was only 4 months ago that Open AI made its GPT-4 model available, for which numerous plug-ins are now available (of which more in future blogs).
But where do you start? What skills – if any – do you need to use Chat GPT?
Because its user interface is so intuitive, some users have input fairly basic text commands (called prompts) and been disappointed with the bland output or even received fabricated information (known as hallucinations). This has led one expert to say “The single most important skill you need is in two words: prompt engineering”, which is essentially the skill of refining and inputting text commands.
While ChatGPT can generate impressive content, merely using it without a proper understanding of prompts, prompt sequences and prompt enginering won’t generate useful output.
So let’s define what these concepts are, and how to apply them effectively:
Prompts: a prompt is a text command you provide to an AI model like Chat GPT as a starting point for the model to generate content based on your requirments. The quality and specifity of the prompt can significantly impact the relevance, accuracy and effectiveness of the AI generated content whereas an unclear or poorly constructed prompt might lead to content that is unhelpful or even misleading.
Prompt sequences: a prompt sequence is a series of related prompts that guide the AI model through more complex and multi-step content generation specific to your requirments. Using prompt sequences enables the user to maintain better control over the flow and structure of the generated content. A useful technique in prompt sequencing is to ask the AI model for feedback such as: Do you understand?
Prompt engineering: prompt engineering is the art of crafting effective and precise prompts to obtain the best possible output from AI models. The process involves understanding how the model works, experimenting with different prompt structures and iterating the input to achive the desired result.
A useful tip when constructing a prompt sequence (a single prompt on its own is seldom enough) is to imagine you are briefing an intelligent new graduate, intern or executive assistant who lacks experience and so needs specific guidance. This may involve giving commands about what you do not want, as well as what you do want.
A well crafted prompt can also stop the AI model from hallucinating by providing constraints that will enable it to create more accurate and relevant results.
Prompt engineering will quickly become a skill that everyone will need to learn; fortunately it is possible to learn the basics in a couple of hours and there are numerous online training courses for beginners – several of which are free – such as this one from Datacamp.
Not surprisingly crafting prompts has become a bit of a mini- industry as budding engineers have created thousand of prompts that they sell online: be careful if you are tempted by these as they tend to be fairly generic and non-specific.
I have been experimenting with various AI models, and in particular Chat GPT, for the past few months and in my experience it is well worth investing the time to craft a prompt sequence that breaks down the steps you want the model to follow, checks for understanding, asks for example output and then refines the prompt sequence to optimise the output. The best sequences will take some time to craft and some of mine run to numerous pages of A4 text; this may seem over-the-top but it has significantly improved the quality and usability of the content generated.