Intro to Prompt Engineering with ChatGPT

So the idea is you can do follow-up messages. When we say prompt engineering, prompt engineering is just a fancy term for texting with a chatbot. Prompts are what we type in; it's just messages.

When you're messaging your friends, you're a prompt engineer. I know that sounds silly, but that's all it is. Prompt engineering is about how to write a good message that will get a good response. There's a human way to do that too, because I'm sure you can send messages that will annoy somebody and not get the response you want.

Same thing with an AI. You can write something and not get the response you want. So in its initial response, if you weren't accurate enough, it's not going to give you the response you want.

So for example, if I go back to my chat here, in this thread here, you know, I said, tell me 20 things you can do. And then later, you know, I said, make it longer. Then later I said, tell me 20 ways you can help me with marketing, right? Because maybe I was thinking, oh, well, I'm a marketer, I know that.

But did ChatGPT know that? Don't assume AI can read your mind. And so at least I, and I've seen a lot of people make this mistake, they go in thinking that AI will just know it all and can give you the answer without you having to explain yourself well enough. And so people often give it too short of a prompt or too short of a message.

They're not specific enough about what you want to see specifically. So the less you type, the less information you're giving it. So it has to make up the context.

It has to add its own thoughts. You know, if you just said, tell me things you can do. Well, how many things is it going to tell you? How long of an explanation? Is it going to be easy to understand? Harder to understand? You know, so you could say, tell me 20 things you can do.

That would help a high schooler. Tell me 20 things that would help a college student. So this is where understanding context is very different.

And so if you forget to do that in your initial prompt, one of the ways is just to follow it up with it. So you could say, give me a simpler explanation. Tell me more about something.

Because you might read something. You're like, oh, let me dive into that more. And by staying in the same conversation, when we think about context, it remembers this conversation.

If you're chatting with somebody in a text message thread, you know, they can look back at the previous conversation and see what you've been talking about. If you just send somebody, some random person a message and say, are we on for tonight? They'll be like, what are you talking about? Because they don't have the context. So this is where, if you start a new conversation, you start a new chat.

It's normally not going to remember the stuff from the previous chats. So that's why we go back and we continue the same chat if we're on the same topic.

So you might say, explain more about, tell me more about analyzing consumer behavior, because you want to dive deeper. So tell me more about, and it understands in this context that we're talking about marketing, right? Because analyzing consumer behavior from a different context would be different, right? This is under the marketing context. So memories are stored across chats.

In the paid version, there are memories and it will say memory updated. And so sometimes when you say like, I do marketing, or when you say about marketing, it might say, oh, this person does marketing and it saves a memory and says, "Oh, this person's into marketing." Or you can even tell it, "I'm into marketing, " and it'll update a memory, and memories are used across chats.

That's a paid feature, not a free feature. If I create a new chat here and I say, tell me more about, it doesn't necessarily know what context that's in. This is general stuff about just understanding consumers, not necessarily marketing, although that's one of the things. So it doesn't, it lost that context.

It has no memory from that other chat. These are completely separate, unrelated chats. It doesn't know anything from one chat to the other because we don't have memories saved.

And we'll talk more about memories in a little bit here. But even in my paid account, because it didn't save a memory, if I create a new chat, it wouldn't have that. But if I say like, I am a marketing person, just since you were asking about it, this is in my paid account, memory updated.

Works in marketing. Now across my chats, it's now learned, "Oh, this is a marketing person." So I'm going to tailor my responses to a marketing person.

And this is why, going back to your question earlier about can it store answers? It can't, because now it has more things to remember. Because now every answer from now on in new chats, it's going to say, this is a marketing person and I need to tailor my responses to a marketing person. So they're going to ask me a question, and I'm going to tailor it to marketing because they're in marketing. That's a paid feature.

Now, if I don't want it to know that, I can always go in and I can manage my memories so I can see these are past memories that I did in a previous class. If, let's say, I was demonstrating this, I could just say, forget that, and I can get rid of those memories if I don't want it to. Or maybe it accidentally created a memory, but memories are a paid feature.

So I think as we go through today, you'll start to realize whether or not you want to pay for this. And while we're just on the topic, we'll come back to talk more about memories again. And it's just saying that there was a TikTok trend of people saying, ask ChatGPT to give you a personalized—like, knowing all of what you know about me, assuming you have a paid account with memories—knowing everything you know about me, give me a plan of how I can change my life for the better.

And it would tell them based on what it knew, and the more it knew about you, the better it could give you specifically a plan to make your business better or a plan to improve your life, like what are the things you need to work on and make yourself better, you know, and it would give them a plan because it has knowledge about you. The alternative is if it doesn't have those memories, you could also tell it those things, but in every chat that you do, you'd have to remind it, I'm a marketing person. So you always have to type in, I'm a marketing person. Whereas if it has a memory, it would always use that.

And you don't have to keep telling it. So do you need memories? Well, you can always tell it in every single chat that you do about marketing, but then you have to type that in every single chat. So the memories help save some time and help save some typing because it can learn more about you.

So normally speaking, we create a new chat when we work on a new topic, when you want it to forget context from one thing. So let's say you're working on two different unrelated projects, I would not put them into the same chat because I don't want it mixing up concepts from one to the next. So I want to create a new chat.

So that has new context and it kind of forgets the content from the other one. But if you want to stay on topic, you stay in the current chat, go back to your previous chat that you have, and just ask it follow-up questions so you can keep enhancing it and improving it as you go through. If you don't like the response, you can ask it follow-up questions. There are also other ways—you can ask it the follow-up question. There's also a way that you can switch models, which we're going to talk about in a little bit, but there's a try again option.

So let's say here in my free account, I didn’t like this answer. See how there's a switch model button? I could say, try it again.

If I just didn’t like that response, I could say, try it again. And it'll just generate based on the same thing that I just said—it'll generate a new response. You might think, I didn’t really like that response.

Can you try it again? Because remember every time you ask it a question, in theory, it could give you a different response, so you could just be like, yeah, I didn’t like that response, try that again and come back to me with something else. If you think that the initial question was good, you just didn’t like the response, you try it again. If you realize you didn’t give it enough context, you want to say something else, do a follow-up message. But if you think your original question was good, you liked what it did, you can go in and you could do it—you can do it again and again. And look, it keeps a history. So, because remember you can’t have it regenerate the same response again.

So what if you realize after you said to try it again, you’re like, oh no, that’s even worse, I want to go back. I want to go back to the first one. There’s a little thing that lets you go back, and so you can go back and forth between the responses.

So it still keeps a history of what you’ve done for that particular response. That’s pretty cool. So you can do follow-up questions.

You can say, try it again. We’ll talk about the other models in a little bit, because I want to explain those models first. We’ll come back and talk about those in just a moment.

You can also go back over your original message. Yes, I know you can follow up with additional messages, but just think about it—if you have a really long stream, it’s a lot to look through. And if you realize when you get a response, you realize, you know, that was a bad question.

I just really want to go back and fix it. You don’t have to just re-ask it and still keep that there. You can actually go back and there’s a little edit button when you hover over.

So, you know, I could say back there, tell me more about how to analyze consumer behavior in three paragraphs. And so now it’s going to use that to generate the response. So I don’t still have to have the previous response in my thread, because the threads can get really long.

And it still does keep the previous question. So there’s still the previous question. If I realized, oh wait, that was even worse—

So you still can go back. This was my first question. This is my second question, but this way you don’t clutter up your chat thread.

Because maybe they like both ones. Right? And this is where it’s so different because you choose the format you want.

Now you can choose the length. You know, if you want bullet points, if you want paragraphs, you could be specific about that. I could even go back and say in 150 words if I don’t want to read too much, or if you need an article written for you—because it’s writing an article basically for you—in a specific word count.

Now, of course, AI is not always the best at writing. Maybe it’s not written the way you would like it to, but this is in 150 words, right? So you could choose how long something’s going to be. And if you just realized, oh, you forgot 150 words—

I could just say, make that shorter to 150 words, make it longer, or whatever, or just change that to 150 words. You could do follow-ups if you want to keep that in your chat, but if you don’t like it to have message after message after message—which does take up more space in the chat—you can just go back and change it, but you still keep your previous ones, which is good. Whenever we open new chats, there are some keyboard shortcuts, for example, Command + Shift + O for open, or on Windows, CTRL + Shift + O will open up a new chat.

So Command + Shift + O will open up a new chat, start typing. And if you forget those or want to learn other keyboard shortcuts, you can click the little question mark down here. There aren’t too many keyboard shortcuts, but there are a couple of them.

So opening up a new chat is something you do a lot. So it’s nice to know the keyboard shortcuts for that.