While some might argue that generative AI is eliminating the need for certain jobs, it’s also increasing the need for new roles and skills such as ‘prompt engineering’.
With many people looking to upskill in this area to produce better results from AI tools like ChatGPT, and some companies creating new roles to stay ahead of AI’s fast-paced developments, we spoke to Mike Loukides, vice president of content strategy for O’Reilly Media, to find out more about prompt engineering and why it’s important.
BN: What will be the role of prompt engineering in the workplace, particularly as the use of generative AI increases within enterprises?
ML: AI is already being used very widely, almost universally, in most companies. Being able to write effective prompts that give you professional-quality output is a skill that almost everyone will need. The ability to write effective prompts is what I call ‘prompt engineering.’ That kind of ability will become a necessary part of everyone’s skill set. It’s definitely something that can be learned.
Prompt engineering can also mean writing software that generates prompts to query AI systems automatically. We’ve seen ‘Prompt engineer’ appear in job postings, and it will no doubt become a job title associated with this specific role. It’s important to keep this distinction in mind: anyone using AI (which will be almost everyone) will need to understand prompt engineering, but there will be relatively few people with the job title ‘prompt engineer.’
BN: Given the threat AI poses to certain jobs, how essential a skill will prompt engineering become to companies in the future or is it just a product of the currently over-inflated state of the AI hype cycle?
ML: I resist the idea that AI threatens jobs. It will threaten the jobs of people who can’t adapt to new ways of working. It will threaten employees of companies that would rather reduce expenses than make better products. The real importance of AI isn’t that you can do things faster, but that you can do things better.
With that in mind, AI is a tool that will enable everyone to work smarter, regardless of what they do. And prompt engineering, in one form or another, will enable people to use AI effectively. I don’t see it as a ‘job’ as such; I see it as a skill that will be essential for many different jobs. Making good use of an AI assistant will become as fundamental as the ability to send email. Creating a prompt engineering group is a mistake that many companies will probably make, but it’s a dead end. Prompt engineering is more like data analysis: yes, there’s a small number of people with the job title ‘data analyst,’ but data analysis is a skill that can be immensely useful to anyone.
A few people have argued that prompt engineering won’t be needed in the future because the language models will become good enough that people don’t need it. That could happen (and OpenAI’s highly anticipated release of their Code Explorer plugin might be a step in that direction), but I’m skeptical. Think about these questions I’m answering. I could have asked ChatGPT to write the answers (I didn’t), but to do a good job I’d have to tell it exactly what points I wanted to make, what tone to use, work through several rewrites, and so forth. Now think about what it would take for me to get a human to write the answers. I’d have to tell them exactly what points I wanted to make, what tone to use… Exactly the same thing. Prompt engineering will only become unnecessary if AI fails, or if AI succeeds so fantastically that it can read my mind. Neither is likely.
BN: What effect will the increased use of generative AI and LLM tools have on companies as they implement these processes into their workflows?
ML: We’re still in the early days of large language models, but we are beginning to see how they’re playing out in the real world. Even if they don’t get much better than GPT-4, it’s clear that they’re going to be important tools for anyone who deals with content, whether that content is words, pictures, or source code. This will affect all parts of any business. That means everyone will need to learn how to work effectively with tools like ChatGPT and, specifically, learn how to write effective prompts. With generative AI touching so many business functions, nearly every employee will benefit from some training in prompt engineering to know what inputs are needed to get the best results possible. That is how employees can prove their worth in an age where AI is quickly advancing.
BN: Given that AI is something that is impacting nearly all organizations, what can companies do to help their employees upskill in LLM tools?
ML: Companies that are serious about building out their technical team must invest in training to thrive in an increasingly AI-driven world. Almost everyone has the same first impression of ChatGPT, regardless of what you do with it: ‘Wow, this is easy.’ A simple one-line prompt will get you a response that looks good, especially if you don’t look at it too hard. However, if you really want professional quality output and not a quick proof of concept, you have to go a lot further. To write a short essay, you will need to tell ChatGPT in detail the style you want to write in, what points you want it to make, what opposing arguments you want to address, and how you want to address them. You probably need to include relevant data, articles, and other outside material in your prompt; ChatGPT was only trained on data through November 2021. You must check its output very carefully; ChatGPT is prone to making up facts and sources, and mistakes here could easily cost you your job. It’s also very good at sounding convincing. Training your staff in writing good prompts, in finding the data you need to help ChatGPT tell your story, and in analyzing its output critically to detect mistakes will enable them to use ChatGPT and other large language models effectively — and it will be much faster (and less risky) than letting them figure it out on their own. Training resources are available; companies must commit to provide those resources.
BN: Are there additional skills beyond prompt engineering that employees need to better utilize generative AI within the enterprise?
ML: Prompt engineering is only the start. Detecting errors is crucial, whether those errors are subtle mistakes in logic or outright hallucination. This skill often falls under the rubric of ‘critical thinking,’ but I increasingly think there’s something more. You’re not debating with an AI; you’re trying to get it to say what you want it to say, and that’s difficult if you don’t know what you want to say in the first place. AIs are very good at being convincing. How do you turn off your built-in will to believe so that you can take a hard look at the facts, the logic, and the conclusions?
You will need to be good at research–not just for checking ChatGPT’s ‘facts,’ but also for supplying it with the facts it needs. Again, ChatGPT’s training data ends in late 2021. (Other AI systems will have different limitations.) It doesn’t have any of your company’s internal documents. If you want it to write about your new products, or your financial projections, or your environmental plans, you have to provide it with the documentation it needs. There are ways to automate this process, but it’s still important to understand what the process is, and how to do it yourself.
Another kind of training will become increasingly important. If AI succeeds in automating many of the tasks that take up our day — even if it only saves 20 percent of our time (which I think likely) — it will be giving us time back. What will we do with that time? The problems we need to solve will move higher up the stack. For a programmer, that means spending more time thinking about the problem to be solved, and less time writing code. For an accountant, that may mean spending more time doing long-term financial planning, and less time creating spreadsheets. For sales, that may mean spending more time researching new customers. I can’t forecast all the kinds of training that will be needed, but I do know that AI will force employees to move ‘up the stack,’ spending more time solving the real problems behind their day-to-day work, and less time on relatively mechanical tasks.
This content was originally published here.
