Prompt Engineering Basics with ChatGPT

So, so far what we've been doing, officially, if you want to sound cool, cool and highfalutin is we're Prompt Engineers. We're writing chat messages. That's all it is.

We're just learning how to write chat messages. But there is something to be said for writing a good prompt, right? All these messages that we're writing are just called prompts. And so there is something to that.

It's not something that just everyone will craft the best prompt or chat message to get what they want to get. Especially when people first get started, they give it too little information. They don't know all the things you can ask it.

And they kind of give it orders. They don't ask for it to get involved. I remember before when I said, ask me any questions you need to do this job the right way.

To engage with it, this requires a different way of thinking. It's not like we go to Google and just ask Google questions. This is a very different approach to it.

So this is all called Prompt Engineering. So there are things that are called prompt frameworks. All a framework is, is just a method for you to remember what to put into a prompt.

So a framework is just saying, hey, let's hear some ideas of what could create a good prompt. And they normally give them some sort of abbreviation because there's probably going to be things that it stands for. As just one example, there's the RICE framework.

Role, Instruction, Context, Constraints, and Examples. These are things that you could think about including in a prompt to get a good response. As an example of the RICE framework.

Role, from the perspective of a Blog Writer. Instruction, what do you want them to do? Write an article about why people should learn ChatGPT to be competitive in today's business world. So it understands, because remember, the more context it understands, the better job it can do.

So this is just a reminder for us to tell it the role. Are we a Marketing Person? Are we a Blog Writer? What's our role? What do we want it to do? Is there any context it should understand? In this example, since it's writing an article, the context is the target audience is adults who either work in companies or are considering changing careers. Are there any constraints or things that it needs to stay within the guidelines? These are guidelines, essentially.

We want it to keep the article between 1,000 and 2,500 words. We want it to use headings to make the article easy to scan and make it friendly for search engines. Any examples? So going back to the question we had earlier, could we get it to write in our style? Are there any examples of either writing that we like that somebody else wrote or examples of writing that we wrote? Right.

So maybe we're telling it to look at a certain structure to say, here's an example of a writing style that I like. Or here's an example of my writing that I'd like you to match. Or I like the way that this article is structured as an example.

Structure this article in a similar fashion. Those are optional, of course, but clients often will say, or you might find things that say, I like this. I want to do something like this.

You could give it examples, and then you can try to do that thing because you give it an example. Now, if you don't have examples, of course, it can use its own reasoning to figure out what to do. But the more information you give it, the more specific it gives you exactly what you want it to do.

If I do that. Let's say I give it this article here. So give it this.

I'm not going to give it any examples here. I'll just do this. And so the idea of the RICE framework is just so you remember.

Oh, RICE. Role, Instruction, Context, Constraints, Examples. It's trying to give you an acronym to try to remember.

And by giving it more information, you are getting it in a format that is more like what you want. You know, I said a certain size of article. I said who the target audience is.

You know, I said from the perspective of a Blog Writer, you know, to use headings, make it easy to scan. So they've got some nice bullet points here. This looks like a blog post that I want to post.

And what would be my next step? I would read the article and make sure it sounds good and that it's correct information. Because could it have completely elucidated certain things? Yes. Also, if this is something that I want to collaboratively work with ChatGPT to do, I probably should have used the canvas.

Because then I could have gone back into certain parts of it and said, well, can you rewrite this or can you change this? I could have bolded words and change words. So. So.

So after this is done, I can switch into canvas. Yeah. Yeah.

And now it's going to redo that in the canvas. Although it did redo it in canvas. If you know something from the beginning, do it from the beginning.

So it doesn't have to redo it in canvas because technically it's a different model. Right. So, unfortunately, you can't put that into the canvas.

You can just say redo it with canvas. Also, the other thing about canvas, too, is down here at the bottom. Right.

There's some suggested edits. So I could say adjust the length and I could say, like, make it longer or make it shorter. Or I can say reading level to make it simpler or make it more, which is basically me just saying, change the reading level to this.

But they're trying to give you some interface things. Part of, I think, also, ChatGPT's challenge is not everybody knows what they can say. Because they can say anything.

So they're trying to give people ideas of like, hey, you can adjust the length. You can adjust the reading level, you know, adding final polish. I guess that just means pour back over it and do like a final shine, make it look a little nicer.

Double-check it and see if there's anything that doesn't make sense. So it'll go back over it and just see, like, maybe as it reread itself, because remember that the normal four hours, it just starts right. It doesn't like right at all.

Double-check itself. The reasoning models do that. The reasoning models like they'll do something.

Check it over. Evaluate. Is this the best thing? Like, they're more reasoning models, whereas four is just like, I just get started.

And if you have to go back and look, maybe it'll go back and realize, hey, I could have improved this. And that's like when I look at the whole thing as a total. Also, if something's taking a while and you don't want to wait, you can also just hit the cancel button.

But it is going back over and thinking. But if you don't have the time to wait, you can just hit stop. I polished the text, ensuring consistency, clarity, proper grammar throughout.

I think it's now ready to use. I'll totally believe it without proofreading. No, I'm just kidding.

Proofread it. Adjust. Oh, add emojis.

Oh, yeah, we need emojis. Is that emojis? I had to have a whole button for that. Emojify.

That's a lot of emojis. See, now, if I wanted more control, I could have sent a chat message. It could have said add emojis to the headings because that that's a little that's a little like I'm sorry.

That's too much. Let me let me cancel that. Add emojis to the headings.

Oh, no, it's still. See. It's hallucinating.

I said the headings, but it didn't do just the headings. It's doing too much. So I want to stop that.

So I could try. I could say add emojis to only the main headings. And I like that you're seeing this because these are the real-life things.

Oh, see now there. Yeah. OK, so.

OK, so it's still working there, but. Where'd they go? Oh, there's another failure again. But it was working.

So my point of this is if you don't succeed at first, try different wording because maybe didn't quite understand. And sometimes you just can't get it to do exactly what you want it to do because it's just truly doesn't understand what you're saying. But maybe try different words.

Be more specific. Try something again. You know, can't hurt to try it again.

One more time. I think it's going to probably fail to edit. But that that prompt did understand.

So I mean, it could finish. The canvas feature is still a little buggy. Here I go to as an example.

Do you put it in the. Do you put it in the message that you can either paste them if you want to give it an article, for example, you can either paste it in. Or if you attach the file, you can say I've attached an article as an example of my writing or format that I like or whatever you want to use to describe it.

You can say, you know, refer to the attached file as an example of blah, blah, blah. Yeah. Yeah.

But you can attach it. You know how that works. Failed again.

So, clearly, we're seeing at least today that the canvas is still a little buggy, but that is why they call it a beta is not finished yet. So, welcome back. Everybody had a good lunch break.

So, moving on here, here's a link. If you want to go check out some other frameworks for additional frameworks again. Excuse me.

All the idea of frameworks, and whenever you hear a framework is, they're just ways to remember things to say to any. So, anytime you learn a framework, just let's try to remember certain things. Along the same idea, the more explicit you are, the more specific you are, the better results you'll get and frameworks are just a way for you to try to remember those things.

So, it's not like there's something special about them. They're just trying to help you be specific and trying to help you remember what things to say. When you're getting into things, it's just really important to try to experiment with different words and things as we said before.

When you're writing different language, if you're not getting the thing that you want, try something different and just try to talk like a normal person to the AI and see what it comes back with.