Podcast: Prompt Engineering: Latest Meme or The Next Big Thing?

In this episode, we discuss how the popularity of generative AI has led to the emergence of prompt engineering which begs the question: what is prompt engineering? By diving into the requirements to use a generative AI model effectively - and understanding the desired outcomes - we piece together the importance of this novel role as generative AI models become more integrated with our daily workflows.  

This podcast is sponsored by Mouser Electronics.  

(6:38) - Prompt Engineering For Generative AI 📝🤖

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn about how students are abusing generative AI models for school work and the potential solutions to mitigate this.

Transcript

What's going on, folks? Welcome back to the Next Byte Podcast. Have you been using generative AI recently? I'm talking like chatGPT, things like that. Have you been worried about how people might be using this thing in the wrong way and maybe thinking about how you can use it in the right way? Or have you considered if prompt engineering is just a meme? Well, you can get the answers to all of that in this episode, so let's get into it.

I'm Daniel, and I'm Farbod. And this is the NextByte Podcast. Every week, we explore interesting and impactful tech and engineering content from Wevolver.com and deliver it to you in bite sized episodes that are easy to understand, regardless of your background. 

Farbod: Alright folks, as you heard, we're going to be talking about AI, GPT, prompt engineering in this episode. So, it is very fitting that we're going to be talking very quickly before the article about another article written by today's sponsor, Mouser Electronics. Now we've talked about Mouser before, but if you don't know already, Mouser is one of the world's biggest electronic suppliers, meaning they got the hookups with all the different folks in industry and various industries, in fact. And that gives them insight into various new leading-edge technologies, AI being one of them. So, they wrote this article, given the hype of everything generated by AI, chatGPT from OpenAI has been just taking off like a rocket. They wrote this article about, okay, the name of the article is actually great. It's called, This Was Not Written by AI. They wrote about the problem of, okay, now we have access to these LLMs, everyone can interact with them. To be blunt, as a student, you can kind of just put in your homework assignment and it'll give you the answers or write you a paper or whatever you want. And that is not ideal because as a student, you're supposed to be learning how to write, you're supposed to be learning how to do your assignments. They're there for a purpose to teach you to train you and if you're just offloading your work onto a model that's already trained by people who did the work before you, that might not make you such a robust individual as you try to move up through your education and eventually get ready for the workforce and all that jazz.

Daniel: Well, I certainly think at some point, right, once you've demonstrated your own level of proficiency, I don't have an issue at all with students using, you know, large language models, generative AI tools as like a way of helping them accelerate their way through their work. I think there's a two-part problem to that, one is that a student claims generative AI work is their own. That's like basically plagiarism and against your honor code. But also, two, if you haven't yet built your base proficiency in writing, you may never ever learn how to write if you've been using something like GPT as a crutch. And I definitely think there's a problem there. I think of my younger sister, she's in seventh grade and I want to make sure that she like develops good communication skills, good writing skills, good critical thinking skills and the temptation to use like a shortcut like AI, that definitely could hinder her learning experience, right?

Farbod: I'm with you. And the way I think about this, I don't know if it's like the best analogy, but it's the one that comes to mind because we're both engineers. Math was like integral to our development as engineers. And I remember like when I came to the United States using calculators was like very, very heavily implemented. And I don't mind that. But I feel like it has to come after you understand, like for examples, for like derivatives or integrals, once you understand the principles of how to do them by hand and you get it, then using calculators makes sense because you get to alleviate a very monotonous part of your mathematical process. And you put more of your mental energy towards like reducing the complexity of the problem that's given to you, right? So, I feel the same way about all these AI tools. And I completely understand what this article is coming from. But it does leave I guess the education world in a pickle because what do you do now? Like how do you counter this? And that's what they were trying to talk about. And they were like, okay, look, there's other AI models that can detect if someone used AI. They give two examples GPT Zero X and DetectGPT. And the other approach could be restructuring your assignments to be so niche and personal. And so, project based that AI won't be able to give a sufficient enough result for what you're going for. Now, I fortunately, I guess, grew up in an age where we didn't have access to LM models and never used AI for literally any homework assignments, even in college. But one of the things that was interesting to me, like as I'm scrolling through Twitter and Reddit, people are like, I wrote my paper, and it got flagged as written by AI. And the only way I was able to like talk myself out of it is by like going to the professor and like showing them the, I don't know, the date stamps of the document and everything. So, I don't know how accurate it is to use another AI model to detect if like a piece of work was written by AI or not that can lead to its own complications. But I really like the idea of, I don't know, targeting assignments to be more specific instead of so like broad strokes. What are your thoughts on this?

Daniel: No, I’m with you, right. So, I’ve heard similar anecdotes that GPT Zero X or DetectGPT are pretty good at determining whether something was written by an AI model. I don't think they're a silver bullet. So, the true solution here is to, for instructors to restructure the assignments in a way that like you're doing project-based learning where it's not centered around like just here's a book and write an essay about it. Or like you said, like making niche or personal topics that like only that person can write about themselves. And then again, that helps someone build a foundation, those basic skills to understand and learn these topics, you know, like learning to do math without a calculator. And then after that, you can turn them loose to all the tools that are available in the world and show them, you know, how efficient they can work now that they've understand and learned how to write on their own.

Farbod: Yep. It looks like we're on the same page. If you guys are interested in digging deeper, by the way, into this topic, we've linked the Mouser article to this episode. In the show notes, you can always find it. But it's a great segue for the topic we're talking about today, which is prompt engineering for generative AI. Now, I'm gonna preface this by saying that, I think a couple of weeks ago, Daniel, I sent you a screenshot of a LinkedIn job posting that said prompt engineer. And I think the caption I accompanied that screenshot with was I thought prompt engineering wasn't me. I thought it was a joke. I did not know.

Daniel: Me too.

Farbod: I did not know that it was an actual job. I thought it was like this thing that we were just all in on a joke together. No, it is an actual profession now.

Daniel: Well, and I similarly to get a gut check, I posted a poll on my Twitter a couple days ago, preparing to record this episode. I said, Hey everyone, I'm just checking up on prompt engineering before an upcoming podcast episode. Want to get a quick pulse check from the community is prompt engineering a real job or just a meme?” 33% of people said it was the real deal. Everyone else said it's as, said it's a meme, right? So, everyone thinks prompt engineering isn't a real job, but we see these job postings. And for people who maybe aren't so up to speed with this about what a prompt engineer could possibly be, right? We're seeing job postings out here for people as their full-time job to carefully curate prompts for generative AI models to make sure that you get the proper outputs. So imagine instead of being someone who writes code, instead of being a software engineer, you would instead maybe be a prompt engineer and what you do is you type in the requests to the AI model and it does the coding for you, or instead of being a graphic designer, you type in the requests to a generative AI model and it creates the art for you, or instead of being a writer, you type in the requests to a large language model and it writes the text for you. Again, you can see why some people might think this is a joke because the AI is doing all the work, right? But again, I like your disclaimer here. I came into this as well thinking prompt engineering is just a meme. It's not a real job, but we saw a bunch of job postings, some at really reputable companies.

Farbod: Some at OpenAI, like that is the company that's leading the charge, at least from what we're seeing.

Daniel: And even roles in the United States federal government for prompt engineering. Like it's a real deal. So, let's talk about prompt engineering. And I doing research for this article, I will say, or for this episode, I will say as highly legitimized in my mind, the role of a prompt engineer. And it's something that I went from saying, Hey, is this a joke to actually being pretty bullish on like the career outlook for someone who might be interested in being a prompt engineer?

Farbod: I'm totally with you. Before we do that though, real quick. I think it's worth talking about like what led to the rise of the prompt engineer role. So, like, what we are, what we've been talking about all this episode is generative AI, right? That's like AI that can look at a set of data and build some level of knowledge about that data and then create something from it. But what it's deriving from that data. What we had before for the most part when it came to AI was discriminative AI, which is, here’s a set of data, distinguish where barriers are, and then categorize things into it, right? So like, if I show you a duck, you're like 99% sure that it is a bird of some sort and not a cow, right? And that's based on the data that you've been trained on. That didn't really require much prompting from anyone. You just got the thing, you categorized it. Generated AI is a different piece.

Daniel: It's still pretty impressive though. Absolutely. Incredible applications in medical technology, but like you said, generative AI is completely different, right? It's not trying to categorize or make inferences about a known set of data. It's understanding a known set of data and then going and creating new things, which again, it's like, it's like trying to mimic human creativity, but with an artificial intelligence model.

Farbod: Absolutely. And I think like where you see it on the application level is what has led it to become such a, I think one of the best adopted technologies even in comparison to the smartphone, which we thought had the best adoption rate ever. And that's because with the discriminative approach, it makes sense, like you're saying, in professional industries, in computer vision for self-driving and things like that. Your average person wasn't using these things. But with generative AI, I can go on there and say, can you come up with a new Persian recipe based on my favorite Persian foods? That interaction is so different now, because you get to garner new information. So now we get into the prompt engineer. Now you have this super powerful tool at your disposal, right? But you need to understand how to talk to it. You need to understand how to leverage it properly. And that's the role of the prompt engineer really, right?

Daniel: Well, and let's talk about why it's needed and the scale at which it will be needed, right? Even though generative AI is pretty darn good. It's really impressive. It continues to impress me, even though I continue to use it on a weekly basis, right? AI writing a story, creating a piece of art, generating a new recipe, composing a piece of music. While that's all very impressive what it can do, there are still inaccuracies in what it does. There are still ways that it can misunderstand what you're requesting. There are still ways that it can be very biased in the outputs that it has based on what the data was trained on. Right? So, being an effective prompter, knowing how to ask the generative AI to create exactly what it is that you want, that's very, very crucial. That's what's created this space for prompt engineers is the fact that if I ask a generative AI model like mid journey, let's say, to create a certain piece of art that I liked, I made one for Nellie the other day so she could use it as her wallpaper on her work computer desktop. It's like, she loves moose and she loves mountains and she loves flowers. I was like, oh, can you like make a moose that's standing in the water at the foot of some mountains with some pink flowers around there? In the style of like, I take it with a photograph, you know, kind of photo realistic, 4k, with the sunset in the back. It took me almost 20 minutes of tweaking different levers, changing my prompts slightly telling mid journey that I like this version of it and not that one and asking it to recreate it and upscale it, et cetera. To get the proper image that I wanted. But again, if I was more experienced in prompt engineering maybe I could have asked and it would have provided it for me in the first run or the second run, as opposed to me tooling around with it for 20 straight minutes. An experienced prompt engineer knows how to evoke the proper response from AI right away.

Farbod: And you know what that reminds me of? Like the very early stages of learning a new programming language, right? Like, let's say you're learning HTML, you're like, I want to make this header look like this. You do it and it's like, oh, it's actually not aligned the right way. I want to align it the right way. And then you keep getting this feedback of what your page looks like, what you want it to be, keep tweaking, keep tweaking, keep tweaking, and then you finally get the right thing. These AI models are kind of languages of their own, albeit you're not literally typing code, you're just giving it instructions for what you want it to do. Which in theory is the same thing.

Daniel: Yeah, that's what computer coding is in general, right?

Farbod: Exactly.

Daniel: You’re telling the computer, giving the computer a set of instructions to perform a task. Right. That's all that prompt engineering here is except for the fact that it is generally these large language models, these generative AI models are trained on the human language, as opposed to something that we created for computers. So, they've got the ability to use our own language to communicate with computers. There's still an element of coding that has to happen there. There's still an element of programming that has to happen to get the proper outputs. That's what these prompt engineers are doing.

Farbod: Absolutely. And I want to talk about this paper kind of highlights, what is prompt engineering? Why is it important? We've kind of covered that so far, but then they talk about what makes a good prompt engineer. And I got to be honest, those couple bullet points is kind of what clicked for me of the importance of this role. They talked about how a good prompt engineer understands the outcome they want from the model fully. They're not going in, hoping for the best. They know exactly what they're shooting for. Then they can use clear and concise language. In addition to that, they gotta have good analytical skills to be able to adjust to prompt, you know, as we were talking about, like with programming languages, trial and error, this didn't work, so now how do I need to adjust to make it work better? And then lastly, kind of bundled them together, they should have some understanding of programming logic that like programming mentality will probably be very useful when interacting with these models. And when I put all of that together, I see this vision of someone that's kind of like a project leader of some sort and a developer of some sort being able to combine their mental energy into a single role because of this generative AI model. Does that make sense?

Daniel: Yeah, no, 100% right. You've got, you know, for a long time, you've needed a product manager, maybe in a team of software engineers to help communicate what the expected outcome is. And then the software engineers translate the user stories from the product manager into tasks that they go work on. And then they go try and develop those specific features. And at the end of the day, they all come together and they hope they built a product that works well for the user. In this case, you've got to combine both parts of your brain, right? You've got to combine the analytical part, the part that understands, how to pull the correct levers with the AI model to get the output that you want, but also keep in mind the overall output that's expected and know how to bridge the gap between the two. I think it's a really interesting critical thinking perspective is the fact that you have to keep in mind both the output and the levers that can help get you there. It feels almost like an indie hacker, an independent person, like working on their own, building a product, being the entire team of themselves, I imagine we'll see a lot more one person startups leveraging tools like this. And a key part of that is, you know, someone if they were to create a one-person business that's augmented by AI, is the fact that they'd probably got to be a pretty good prompt engineer.

Farbod: That's exactly what I was about to make. This is why I love doing episodes with you, man. I mean, I don't do it with anyone else, but anyways, that's exactly what I was going to say. Just imagine like the amount of power we were able to give someone by just giving them internet access right now. You can do internet mailing, you can do internet advertising, you can like spin up your own website with wix in seconds, right? We were able to get so much more done. We had so much more at our fingertips. I see this kind of being the same way you now have so much more at your fingertips so many more levers you can pull but you just have to know how to use the tool and that's I don't know if prompt engineer is the best title for what this person does or is, but that's really what I'm seeing out of this role.

Daniel: Yeah, no, I'm with you. It's like, you're more than, at least especially right now in the early stages of this generative AI, right? You're not a code monkey or something like that. That's just like typing stuff in and expecting a different output, right? What you're doing is you're integrating AI into a known workflow. So you're almost like an AI integration engineer just as much as you are trying to specifically engineer a prompt. I think it's super interesting. And again, I just kind of want to reiterate some of the places where we've seen prompt engineering so far start to pop up. This is like real job openings for prompt engineering. If you're looking for a job, maybe this is a career path that, that you could take a look at, especially if you meet some of these qualifications that Farbod mentioned earlier, but we've already seen it in media and entertainment, in customer service, we've seen it in high technology, including open AI, which one of the leading companies in the generative AI space. We've seen prompt engineering jobs in marketing and manufacturing, in defense and in finance. These are all like broad sweeping parts of industry, you know, of the corporate machine. These are big parts of industry that are looking for people to help implement generative AI in ways that can help overhaul, can help change, can help make their businesses more efficient. And being a prompt engineer is kind of the key link between, you know, all this cool technology that's out there and integrating it into the workflows of these companies.

Farbod: Absolutely. And there really is an emphasis on like the co-creation aspect here. I know a lot of people, like at least from what I'm seeing on social media, it's like doomsday's here. Generative AI is going to kill every one of us. This is Terminator. Like these are the final days. But the reality is, is like, where people see the most potential in these models is, hey, I'm going to give you the guardrails. I'm going to tell you what I want and I will guide you so that you can tell me exactly what I need to accomplish that. And in that way, it reminds me a lot of like generative design for mechanical design. Like you and I, that’s exactly where my brain went. Really? OK. Yeah. Again, there's a reason we love doing episodes together. We're basically two halves of the same mind. But I remember working with those generative design tools, like with in school. And I had some level of understanding of like material science, of manufacturing, of like what's a good design and what isn't. But these tools, they give you a lot of different options and they're even customized of like; this is great for 3D printing, this is great for like thermal loads, whatever. And I could look at it and I could choose and I could move forward now for my, you know, senior design project or whatever. It wasn't that important that I got the load right. But I'm imagining for like some engineer at a massive consumer, electronics company, let's say Apple, right? They really want to optimize cost and weight. They would maybe leverage a tool like this and they probably wouldn't give it to some junior engineer. They'd probably give it to someone that is very, very well-versed in the topic of mechanical engineering and manufacturing because they have the insight of what makes the most sense and what kind of boundary conditions to give that tool to get exactly what they want, right? And instead of spending their time doing the meticulous monotonous designing process, they get to just channel their knowledge into this platform and the platform will do the heavy lifting for them.

Daniel: And I like the analogy you used earlier with a calculator, right? The advent of the calculator didn't completely obsolete all scientists and engineers, right? It accelerated it. It accelerated those fields and it allowed people to use their brain for critical thinking rather than solving problems by hand. And I think that there's a very similar trend we'll see here with the other things that generative AI can do is, people will still be a huge part of the, the secret sauce, the formula here to get successful outcomes, right? Is they will be doing the critical thinking and they'll be solving the problems, but they won't have to do a lot of the work by hand to get to the proper outcome. So, I think it will continue to accelerate these fields. We'll see, I'm hoping like a huge productivity increase from people working in these spaces with generative AI. And you mentioned it earlier, how fast this stuff is growing. And I wanna be sensitive to the fact that you and I, as tech evangelists, you know, science geeks, nerds, whatever you wanna call it, right? Like we're probably inside a bubble of people who are really bullish on AI, really bullish on generative AI. I wanna give some statistics for people who maybe haven't had as much exposure to these technologies and kind of, so you can start to grasp the trend that's going on here. There are tons of examples. Chat GPT is probably the most prolific of any of these tools. In the first two months after it launched, it already reached 100 million monthly active users. That's the fastest growth of any consumer application in history. It is projected to make 200 million dollars in revenue in 2023 and $1 billion in revenue by 2024. This is a huge tool, a huge consumer application, and people in all sorts of realms are using it, not just people inside the tech bubble, although certainly Farbod and I, we're probably more in an AI echo chamber than a lot of people who might be listening to the podcast.

Farbod: For sure, but I will say, I've been using ChatGPT for a lot of the work that I do, and I think that's part of the reason why I'm so excited about it. I do a lot of programming in my day job. There was like a great example I could give is something I was stuck on. I did my usual search. It was very niche topic too. Nothing came up. I was beating my head to the wall and I'm like, you know what, let me just try chat GPT. Gave it specifically the conditions I was looking for, the boundaries, the limitations, everything. Within probably 15 to 20 minutes of trial and error, I was able to get exactly the answer I wanted. And that's like, that's so incredible because it unblocked me from what might have been another couple hours of searching and then manually trying to get myself over and over again until I got it right. And that's, that's the beauty of it. It can open up people's time to be so much more productive and effective at the things that they do.

Daniel: And I'll even, I'll come clean here, right? A lot of the podcast episodes we've been doing research for, I've been using GPT as a lever to help me understand, you know, broader amounts of technology, read more or cover more ground in doing my background research before we come and talk about one of these episodes so that I've got a more full understanding. Spending the same amount of time to research. I can cover twice or three times as much ground using GPT as a tool to help me summarize and get the important insights out. You know, I've got a better understanding of the background so that when we have these conversations, hopefully we're delivering a better product. And AI is a big part of that.

Farbod: Absolutely. So, to recap, right. The world until very recently has been mostly using AI for discriminative AI, as in categorize this into this, use it for computer vision, is this a duck, is this a bird, whatever. But the rise of generative AI, which means given all this data that you've been trained on please create something new, has led to well how do we really use this tool to get the best out of it, make sure that it's accurate, and it's the thing that we were originally intending to get out of it made this role called a prompt engineer which until very recently Daniel and I thought it was a meme and according to Daniel's twitter poll about 2 thirds of people had the same idea that it was just a meme. Well in fact it's not. We are seeing job postings for this role in the federal sector and the private sector and some of the biggest companies that are doing anything and everything AI and as it turns out what this article really dives deep into is that a good prompt engineer understands the outcome, can use clear and concise language has great analytical skills so they can keep iterating on their prompts and has hopefully some understanding of programming logic to know how the model is working. That sounds more like a product manager, project manager combined with some sort of a developer or a creative person. So, what we think the future of the prompt engineer is going to be is A, leading to solopreneurs, like people that have ideas, being able to get off the ground much quicker because they have so many more tools at their disposal, if they can get good at this skill set. And then B, making everyone else that adopts this level of proficiency with this tool, so much better at what they already do. How do I do? Is that a good summary? Thank you.

Daniel: And I like the part at the end, right? This is a skill set that people can develop. Even if you're saying, you know, maybe I don't wanna devote my career to being a prompt engineer. Prompt engineering, that skill is almost certainly something that you should develop and maybe even list as a skill on your resume to show that you've got the ability to use these tools to, you know, maybe you're only a 1 or 2 X engineer on your own, but with AI, you could be a 10 X engineer.

Farbod: I can't wait for LinkedIn to have a certified prompt engineer badge that you can display on your profile.

Daniel: I would not be surprised if it comes out soon, man.

Farbod: Okay. And I think we're good to end the episode. Yeah?

Daniel: Yeah, let’s do it.

Farbod: Alright everyone, thank you so much for listening. We hope you liked this as much as we did. And as always, we'll catch you in the next one.

Daniel: Peace.

-------

That's all for today The NextByte Podcast is produced by Wevolver, and to learn more about the topics with discussed today visit Wevolver.com.

If you enjoyed this episode, please review and subscribe, via Apple podcasts Spotify or one of your favorite platforms. I'm Farbod and I'm Daniel. Thank you for listening and we'll see you in the next episode.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!

--

The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.