??? Prompt engineering makes a comeback … sort of
Headlines You Should Know?
Prompt Engineering Makes a Comeback … Sort of
The torrid pace of AI development means there’s no clear playbook for how to use these tools, and best practices can change at the drop of a hat. In the early days (of November 2022), the term “prompt engineering” took off. It was thought that a carefully designed input would produce an optimal output. Then, the guidance changed — it was believed that large language models were so good (and improved often as tech companies like Anthropic and OpenAI rolled out updates) that prompting would become less important.
We’ve seen the start of this with tools like DALL-E turning a simple prompt from the user into an elaborate description of an image, but we’re not getting there as fast as many assumed, and new prompting guides aim to tide us over until then. A paper published Sunday by Ethan and Lilach Mollick of UPenn’s Wharton School of Business focuses on prompts for educators that result in personalized learning experiences for students. Similarly, Google recently published a prompting guide for its AI tool, Gemini, that focuses on four main elements: persona, task, context, and format. OpenAI shared a prompt engineering guide late last year, too.?
With a renewed focus on prompting, communications pros should think about a couple of big things. First, there is no such thing as a perfect prompt. The nature of AI is that it will respond (at least slightly) differently each time, so there is no silver bullet. The elements Google lays out are a good start, but no guarantee you’ll get a terrific output.?
Also, a lot of AI hype is warranted, but it also generates pressure on the user to get the best of the tools. Instead of going for the home run with every prompt, play small ball. You could waste a ton of time tinkering with prompts or re-prompting to get exactly what you want, but you’d get better time value with the singles and doubles, taking a smaller win that saves you some time by sparking a new idea or giving you a good basis to start from that you can finish manually and quickly. That’s the efficiency gain that everyone should aim to achieve.
Elsewhere …
领英推荐
Tips and Tricks
?? How to edit parts of ChatGPT outputs
What’s happening: A few weeks ago, OpenAI added an editor for DALL-E images within the ChatGPT interface. The same functionality is also available for written outputs.
Why it’s important: AI tools tend to start outputs from scratch with each prompt unless you ask for something specific. This capability allows you to select just a portion of the output and ask ChatGPT to make changes. Maybe this means something very detailed, like finding a synonym for one word, or even something a little more broad, like reworking a paragraph to include a proof point or example.?
Try this: Drag your cursor over the part of the output you want to change, and you’ll see a little bubble appear above it with a quotation mark icon. If you click that button, you can re-prompt to change the text you’ve selected. Interestingly, this function also works within GPTs.
Quote of the Week
“Innovation requires cooperation. Nobody knows exactly where AI is useful, and where it is not, in advance. I have yet to see robust communities of educators (or many other professions) coming together to share what they create. Policymakers and institutional leaders can help make that happen. In the meantime, we hope that our prompts and blueprints can serve as the basis for ethical experimentation. Just make sure to share what you learn!”
— Ethan Mollick, author and Professor of Management at Wharton, in his latest blog post
How Was This Newsletter?