Since the boom of ChatGPT at the end of 2022, workers have wondered how much they should use generative artificial intelligence to help assist them in their daily tasks. And if they do, what questions should they ask it? And some have questioned if they should use it at all, since some companies and even countries have banned it entirely.
But it’s becoming increasingly clear that knowing how to use AI will become a must-have skill for workers in the not-so-distant future. That means everyone needs to have a little bit of a prompt engineer in them.
Microsoft’s 2023 Work Trend Index, released this week, included in its top three findings that every employee needs AI aptitude.
Leaders Microsoft surveyed said it’s essential that employees learn when to leverage AI, how to write great prompts, how to evaluate creative work, and how to check for bias. It’s a result of the new way of working, which calls for human-AI collaboration. In fact, as of March 2023, the share of U.S. job postings on LinkedIn mentioning GPT are already up 79% year over year. And 82% of leaders in our survey say their employees will need new skills to be prepared for the growth of AI.
“Using generative AI is like a secret key to the treasure chest of the future of business intelligence,” said Marwan (Mark) Kashef, who is an AI freelancer on Fiverr.
The report encourages leaders to help people embrace a new way of working, starting with building this AI aptitude, which ranges from nailing perfect prompts to fact-checking AI-generated content.
That can be done by leveraging learning resources and crowdsourcing best practices from employees as they begin to use the technology.
For example, web hosting company GoDaddy recently launched a “Prompt Library,” specifically for small businesses looking to take advantage of generative AI’s assistance in certain areas of managing and growing a small business. And getting a grip on how to use generative AI to help with work will not just be an expectation for the technically savvy, but for all workers. But the speed of AI’s current developments can make it hard to know where to start. Prompt libraries like GoDaddy’s can help, telling you where to create an account and it organizes prompts by category.
“Like any new technology, people are all missing one major thing, which is time,” said Kashef. “Time to actually learn the thing.”
That challenge is one that a lot of folks are facing with generative AI. There is no secret that AI can speed up repetitive tasks and free up time for deeper work, but it can be hard to find the time to tackle the learning curve and figure out the best way to use the technology. Users who are not familiar with generative AI might be frustrated when it takes too long to get the right output or they spot hallucinations with the output.
“Computers are still dumb in the sense that they can only follow your exact instructions,” said Haris Butt, head of product strategy and design at project management platform ClickUp. “You can’t assume anything. You have to tell it exactly what to do. It’s the same for prompt engineering.”
John Jung, vp of engineering at workplace automation company Nylas, seconds this: “Why isn’t it listening to me? It hits you like a train when you realize. If you met a stranger that has like 15 [college] degrees, who can be super helpful, and you just said a random thing to it, what is it going to say?”
That’s why prompt engineering is so important. Jung and Christine Spang, co-founder and CTO at Nylas, say the best suggestion is to start by spending time with the chatbots to learn how they work through a trial and error method.
“The idea is to just play around with it and ask it to write formulas or emails for you,” said Jung. “Then try to respond and get it how you want it, like ‘this is too formal, make it more casual. You start providing more instructions and it will do what you want it to.’”
But there are some specific tips that experts say can help speed that process along. For example, Butt recommends using a bulleted list when you prompt the bot. He says to consider questions like how do you want the AI to act? What do you want it to do? What is the output you’re expecting? Kashef suggests integrating as much nuance as possible to the prompt so that you can refine it to really narrow down the output it will give you. Additionally, you can act out a role-play where you tell the bot what it is so that it responds in that way.
“The quality of how smart you’ll get through AI will only match how well you’re able to instruct it based on what you know,” said Butt.
But either way, he says that prompt engineering is “still too abstract for people.” That’s why he says it’s helpful even to change the language from “prompt engineer” to simply just chatting with it. In that way, it humanizes the language and helps more people understand that they can also navigate this technology that is still new to most of us.
“If you can speak the language, just like with any person, if you can speak their language, you can drive the most value out of the conversation,” said Kashef. “The same logic applies.”
Knowing how to speak the language also makes it easier to fact check and verify its outputs, another crucial part of the process. Kashef recommends asking for the bot to return its source.
“Try to ask it to return the source with that piece of content,” said Kashef. “I say ‘hey, give me 10 facts and support it by some web pages or blog posts you’re getting this from.’ It’s trying to make it have its own reference sheet like a student would have for an essay in college.”
He also suggests that if you are asking multiple questions to fact check as you go, instead of just fact checking the last thing you are hoping to get to. Once you learn how to nail prompting, the tool can help immensely when it comes to speeding up everyday tasks. Experts say it will also become less and less of a standout skill because more people will get the grasp of it.
“It’s a hot field, but it’s something that will become less of a skill overtime,” said Jung. “They are constantly tuning these systems and models to make it more accessible and as available as possible. Over time, the hope is that you don’t have to be a trained person to use this.”