When GenAI tools like ChatGPT first arrived we were all amazed at how useful it was, even though we hadn’t had any instruction on how to use it.
Over time we began hearing about prompt engineering, which simply meant learning to frame good questions. [Did you know, the word ‘prompt’ – as in an instruction given to an artificial intelligence program was one of Oxford University Press’s shortlisted ‘Word of the Year’ for 2023’, announced just before Christmas].
But now we are now entering a phase where prompt engineering is getting more sophisticated and HR tech professionals should ponder how to develop the organization’s expertise in creating prompts.
So, let’s spend a moment on basic and advanced prompt engineering, and then consider what it takes to build organizational capability in using prompts:
Basic Prompt Engineering
For those not familiar with prompt engineering, here are a few examples:
- Weak prompt: Tell me about change management
- Better prompt: Tell me about change management needed for a new technology implementation in the aerospace industry.
People have learned that adding detail can lead to better results. We’ve also learned there are many good prompts that we might not have originally thought of, such as:
- “Provide examples of how to do ____”
- “If needed, you can ask me questions to clarify the prompt before you answer?”
- “Respond as if you were the CFO”
Clearly, you’ll want employees to develop their basic prompt engineering skills so that they can get the most out of Generative AI.
Advanced prompt engineering
Recently, we’ve gone a long way beyond basic good prompts to a new level of sophistication.
Experts suggest that instead of thinking of the prompt as a question, you can think of it as the kind of briefing you might give to a consultant.
A briefing might look like this:
– Define the goal
I need a detailed step-by-step plan to convince my CFO that increasing pay will reduce turnover.
– Describe your situation
I’m new in the job and need to visible win. I want to do something to show I’m business savvy.
– Give the context
We have four small but fashionable hotels in Austin, Texas. Our turnover is always about 20% higher than industry norms. Not only does this affect recruitment and training costs; I think it shows up in customer satisfaction.
– Explain what you want the output to be
Can you give me three different ways of approaching this? Show step-by-step in bullet points how I might make the pitch to the CFO.
This kind of advancing prompting not only gives impressive results, it helps us reframe our understanding of just what kind of entity a GenAI is.
It really is NOT just a fancy Google search.
Chain-of-thought prompting
Another advanced method is “Chain-of-thought” prompting, where instead of just asking one question, you take the GenAI through the thinking process, one step at a time – just like you might if you were working with a peer on a complex problem.
Imagine saying to a peer “Okay, let’s work this through one step at a time. First, let’s list what we know about the situation….” And so on.
You can work with a GenAI in exactly the same way.
Few shot prompting
You can also give the GenAI examples of how to do something.
This is called “Few Shot Prompting.”
It helps the GenAI just like it would help a student learning a new subject.
For example: The point of this blog is not to teach advanced prompting, just to make you aware that there are a variety of approaches that go way beyond giving good simple prompts.
Do we need sophisticated prompt engineering?
OK, so that’s the theory, so do we really need ever more sophisticated prompt engineering?
It’s certainly true that most people, most of the time, won’t need to use advanced prompts. This is just like most people don’t need to use advanced features in Excel.
However, for the organization as a whole, lacking knowledge of advanced prompt engineering means being unaware of the true power of GenAI.
As such, you ought to have people who will recognize that for some situations we will get far better outcomes by accessing the full potential of our GenAI tools.
We need communities of practice more than ever
If you want to know more about basic statistics then you can take a stats course or read a textbook because the topic is well-understood.
Prompt engineering is different, however. It’s being invented as we go. Hence, what’s really needed is to keep up to date – and that’s hard to do alone.
Organizations need to enable communities of practice that help groups of motivated people learn from each other and advance the overall capability of the organization in some fields.
The obvious department to spearhead this is learning and development.
They are far better off setting up communities of practice to advance knowledge on evolving subjects like prompt engineering than sending people to some kind of course.
L&D should have the expertise in setting up and supporting communities of practice – and this only requires a light touch. It’s literally something that can be done off the side of the desk.
L&D would normally look to the different departments as to what areas need communities of practice, although, in the case of GenAI, I think it’s an obvious need.
The only caveat to the issue about prompt engineering is that it may be that the GenAI gets better so fast that it can do its own prompt engineering better than humans.
But, until this happens, let’s give human employees the support and encouragement they need to build that expertise and capability internally.
