Exploring Prompt Engineering with AI for Government Employees

Let’s do an activity where we get started using prompts. In my browser here, I'm going to go to chatgpt.com. I'm using ChatGPT in this course because a lot of the agency-specific AI assistants are based on ChatGPT, and even if they're not, the interface is fairly universal. So what you're seeing here, even if you're using a different model like Grok or Gemini or something, a lot of the same principles will apply.

So I'm going to chatgpt.com, and I'm in an incognito browser. I don't even need to log in to get started. And let's ask it a prompt.

What you put in this ask anything little context window here, that is a prompt. So let me ask a very simple question. I'm going to say, tell me 10 things you can do.

Here are 10 things I can do, it says. And very quickly, it's returned to me. I can answer questions, generate creative writing, assist with math and problem solving, coding, translation of languages, recommendations, and make visuals.

I can be a conversational partner, help with planning an organization, and analyze data. So this, what we just asked for, well, that sentence that we put in is considered a prompt. Now we can follow up with that prompt.

So, 10 things you can do. Okay, that's great to know. But I'm, say, a government employee, and I want to know how it can help me as a government employee.

So let's ask it. Tell me 10 things you can do to assist a government employee, sorry, like me. And notice that the additional context that I just gave resulted in a completely different answer.

It still did what I asked; it gave me 10 things. But before it said, you know, I can answer questions, it talked about creative writing, math, and visuals, and things. Now that I specified that I'm a government employee, it says, oh, I can analyze policies and research, I can draft and review your documents, data analysis, which is what it said before, create presentations, that's new, regulatory compliance assistant, task management, public communications and engagement, conflict resolution, negotiation tips, training and capability building, and capacity building rather, and legal research and case studies.

So notice, just that little bit of additional information that I put in my follow-up message to continue the conversation, gave a completely different result. We're starting to see the benefits of prompt engineering at play here. In another tab here, I've logged into ChatGPT using an actual account, and because logging in or creating an account gives me some additional options.

And the reason I did that is I wanted to showcase a certain aspect of interacting with any AI assistant. That, by the way, is a specific term called prompt chaining, where you ask additional questions after your first prompt to get a more specific answer. So you could continue with that same chat, but you could also create a new chat when you log in.

And a reason you might want to do that is anytime that you have a completely different topic or a different question, unrelated to what you're currently talking about, it's a good idea to start a new chat. Though you know, these GPTs are extremely smart. It would probably catch up with you, or probably keep pace with you, and hang with you.

But a good, just kind of hygiene rule is keep conversations distinct from one another. And when you're logged in, a nice feature of ChatGPT and many of the other AI assistants is that it will log those chats. So you can also go back to it.

And so, for all sorts of reasons, continuity of the conversation, for future reference, if you need to go back and refer to some, you know, some of the conversation that you had had with one of your AI assistants, it's a good idea to keep like topics together. And if it's completely new, it's completely different. Whenever you completely change context, it's a good idea to switch to a new chat.

So let's do that. Let's completely change our topic here. It's no longer telling me what it can do to assist me as a government employee.

I've switched topics. I'm now talking about prompt engineering. So I've started a new chat here.

And what I want to demonstrate with this response, when it comes through, and here it is, that was really fast. It says, “Here's a fast, practical cheat sheet.” You want to start with the outcome in mind, load up some content, and show the shape.

You can prompt in building blocks. And we'll talk about this a little bit later in the course. This is really, really useful stuff.

But maybe let's say this is way too long for me. Whoa, whoa, whoa. Why are you giving me code? I don't need a promotional teaser.

I didn't ask for any of this. So we'll let it stop rendering here, and I'll show you some things that you can do. Okay.

It looks like it's completed. We can ask a follow-up question to clarify. So let's say I wanted to ask about this code.

You know, I could say, why are you showing me code? You know, that's one thing I can do if I'm not satisfied with this response. Or if it's way too long, like this arguably is, you could say, "Give me a much shorter version.” You know, that's something you can do.

What I also want to highlight here is that you can ask it to regenerate its response. So notice it gave me this response. I could ask additional questions, but we've already demonstrated that.

So I won't do that here. But another option you have, notice when I hover over this icon here, it says, try again. It tells me what model it used.

In this instance, it was a ChatGPT-5 thinking. But I'm going to ask it to try again. Specifically, yeah, try again.

So notice it erased that, and it's going to take another swing at my prompt here. So it has finished rendering. It thought for a couple of seconds and then generated the new response that I asked it to retry.

And here, I wanted to highlight a couple of things. Down here, because I'm logged in and because I asked it to try again, let's say maybe I like this. This is even worse than the first response, let's say.

I can switch back to that first response. That's always available to me. I can go between any one of the versions.

And I could ask it to try again and again and again. And it would provide those arrows and allow me to cycle through those different responses. So that's certainly something you can do.

If you put in an initial prompt and no follow-up questions are necessary, you can ask it to simply try again. Another thing you can do is maybe you realize I've done this two or three times, and I'm just not getting the result that I was hoping for. Notice when I hover over my initial prompt here, where I asked what everyone should know about prompt engineering, I get a little edit icon.

So I can click this, and I can edit that initial prompt. Maybe it was just a bad prompt. And actually, if you've read any of its advice, you'll know that it is because it doesn't do a lot of things that a good prompt should do.

So let me add at least one or two of those things. So instead of being so vague as I've been here, where I said, you know, what's something that everyone should know about prompt engineering, let me get more specific. I'm going to say, what should government employees know about prompt engineering? And I'm going to add some more specificity to it so they can excel at their work.

And watch how it changes just with a little bit of extra detail here. Ah, thank you, ChatGPT. An excellent and timely question, it says.

And here, as it renders, it says, provide some headings, and understand the purpose of prompt engineering. Bravo for you, you're in this class. Learn the core techniques.

Prioritize data privacy and ethical use, of course. Use prompts to enhance daily functions, build AI literacy, and apply a federal mindset. And then it even says, "Would you like me to create a training module or worksheet for you?” So I wanted to highlight a couple of things here while I was logged in, because there's a lot you can do.

But notice how much it improves with just a little bit more specificity with your prompts.