You Probably Don’t Need Prompt Engineering!

So I keep getting this question over and over again. Igor, is it worth to even learn prompt engineering in 2024? And I get it. I think it's a very, very relevant question. But the answer really depends on what you're trying to accomplish. I'm going to try to break it down in this video. And not just that, I'm going to give you a tool that will solve many, many, if not most of your prompt engineering struggles. But more importantly, I'm going to clarify why you might not even need that tool. Why you might not even need all these fancy prompt engineering tricks. So let's get into it. I think this is going to be a super value packed video and I think you might be pleasantly surprised by my breakdown of if you even need prompt engineering these days. So let's start by establishing what prompt engineering even means. There's many definitions floating around, but I like to see it as getting the best results out of tools like ChatGPT. I think that's what most people mean when they say prompt engineering. It's interacting with LLMs like ChatGPT, Claude, Gemini, Llama 3, etc. To get the best possible results out of it from for themselves. This is what most people mean. Sure, there's prompt engineering in the context of image generation, but that's a different topic. And usually when people bring this up, they mean chatgpt interaction. That's why we will be focusing on that today. So why is this called prompt engineering and not using ChatGPT? Well, because the main interface with ChatGPT is prompts. It's natural language. It's set up for humans. We can speak to it in the same language that me and you both use to communicate in our everyday lives. And then prompt engineering is manipulating that language and being aware of version various tips and tricks and phrases that then improve the results or unlock entirely new abilities altogether. But it's interesting because if we look at the example of ChatGPT, so many tools have been added into it that getting the most out of it goes beyond just knowing what to type into that little chat box. You have image generation, you have data analysis, you can store things into memory. You have GPTs with external knowledge bases and actions that can do things by connecting to external APIs. This goes beyond just telling it the right words to get get better outputs and over time my personal definition when I answer these questions is really all of this. It's a holistic look at ChatGPT, all the tools and capabilities combined. Because for example, to use the data analysis tool, well you need to be aware of the different capabilities of the different types of graphs it can produce for you, and then you can prompt it to do that, which brings it back to prompt engineering. That's why I'm saying I include this in my definition. So now that, at least for the context of this video, we established our definition, do you need prompt engineering in 2024? And the answer is it depends on what you're trying to do. I would break this down into two categories. Okay, we'll keep it simple. The first category is you're interacting with ChatGPT yourself. You're sitting there, you're typing things in, you're getting results, and you have the ability to go back and forth just like you would with a human assistant. Now, for most everyday users, for most consumers, this is the main use case. You sit here, you open a new chat, and you start typing requests, and then you get results, you look at them, and you follow up with more messages. And in this use case, where you're there to do the quality control, to follow up, to include other things, I think most people don't need prompt engineering. A basic education is extremely valuable. Being aware of all the different toolings, being aware of some of the different prompting tricks or some of the hidden abilities, like, for example, the ability to generate interactive quizzes, or knowing that you can just throw Excel sheet at it and also chat with that. And this is what I've been pushing on this YouTube, right? I've been educating you around the tooling about different prompting tricks, but at the end of the day, if I want to write an essay, I don't think most people need super prompts in their everyday usage. And some of the most advanced ChatGPT users I know use it like this. And so do I myself. I have a prompt database with a lot of specialty prompts, but I don't even use it on an everyday basis. If I need something, I just tell it. But having spent quite a bit of time, I have kind of a backlog of different functionalities and prompts in my head that I can kind of access. So in these everyday scenarios where you dare to interact with it, I would say that beyond the basic education, you don't need prompt engineering. You just need intuition and certain quality standards that you will uphold ChatGPT to, because when you review the results, that's where the next prompt kind of comes from. If you're missing something, think, tell it. And this is why I always teach ChatGPT in a way where I tell people to start by thinking of it as a very highly paid assistant. If you hire an assistant for yourself, you have certain expectations and if the results are not on par with what you expect, you'll tell him or her. And then when you get the revision, you'll feedback it again. And this is the usage that even advanced users kind of arrived at over the past few years. And this keeps resurfacing in conversations. Matter of fact, recently I talked to Javi Lopez from Magnifica. You might know them, they're probably the best image upscaler on the market. And I say that objectively, I said that before and he came into the AI Advantage community and he shared his story with us and in the end we had a little Q and A with the community and one of the questions that came up was how do you prompt these days? And this was his answer. I wasn't like, you are an expert in whatever, you are an expert in programming. How is he? No, I'm pretty straightforward. Maybe I think I'm living in the future and maybe it will be better to be more specific or more using prone libraries or whatever. Maybe, and this is my point, if you have a certain level of experience, most of the time you'll be fine by intuitively using ChatGPT because you have the ability to go back and forth and feedback with 32,000 tokens of context available within this interface. Now that's enough for long conversations, so you can do this iterative feedback process. But again, that's where the base knowledge of ChatGPT comes in. You need to know what tokens are. You need to know what a context limit is. By the way, if you want to learn more about this, I created a deep dive lecture on it that I made publicly available in the public area of the AI Advantage community. You can, you can check out a link in the description below, but it's basically a one hour lecture only focusing on token limits, as this is one of the essential things you need to learn about when you start out with ChatGPT. So that is the everyday usage for most users that is also relevant, right? And I just kind of stated you don't really need prompt engineering while having a prompt engineering course. And this channel being mostly focused on prompt engineering. Why do I even bother then? Well, that's because there is a second bucket of usage that is very different. And I say bucket because it includes multiple scenarios. It's basically every scenario where you are not there yourself to review the answers and to apply what you know to the conversations. This includes, but isn't limited to creating prompts for other people. Maybe you're an entrepreneur and you want to equip your employees with various prompts. If this is the case, you won't be there to follow up and to massage in all the little edits. You kind of want to get it right in the first prompt. And in this concrete example, it might make a lot of sense to build something like a GPT. So you give them a simple user interface with presets and a communication sequence that is preset. We've talked about this before. There's many guides on my channel about this. But that's not where the second category ends. Because there's other scenarios where you absolutely do need prompt engineering, and that is when you're dealing with automations. If you automate something, by definition, there's not going to be this human touchpoint, this review process, and that automation, which includes a prompt, is going to run dozens, maybe even hundreds of times and there's no one there to double check the results. In this case, you do want prompt engineering. You do want to flesh out the context and define all the little variables that might cause alterations in your results. You want to make it consistent, right? And that's why I put this in one bucket. Because just like with the entrepreneur crafting prompts for the employee, you want consistency and reliability. And when this is important, prompt engineering is hyper relevant. And there's also this third scenario which belongs to this bucket, which is when you have a use case that you use over and over again, maybe a prompt that you use once every week. This is where you also want to flesh it out, because you want consistent results every single time. Every time, where consistency and predictability is paramount. And engineering your prompts and putting in the extra time to flesh out all the context is probably a good idea. Okay, so now we establish these two buckets, right? One where you interact with it and you can feedback, and one where you won't get to feedback the results. So what to do about the first one we covered already, right? Most of the channel is about that. It's being aware of different prompting techniques and gaining a basic education on what tools exist and what they can do. But for the second bucket, you really do want to get into more intricate techniques. This is where super prompts come in. This is where fleshing out a one line prompt into something that spans across the entire A4 page is probably a good idea. But how do you do this? Because this is a skill that is not that easily taught. It's not something you can learn in two free YouTube videos. And no, this is not going to be a sales pitch for another course or something because even experienced prompt engineers that do this as their full time job need a lot of time to extend these prompts and to flesh out all of the context. But I think we found a solution that will solve 80% of these issues for most people and we've been working on this solution for months and about to give it to you for free. This solution has been co developed between Rastislav Dujava, a Czech prompt engineer that I actually met early 2023 in the AI Advantage Discord community morphed into the AI Advantage community today, but Rastislaw basically came along and we started collaborating on a lot of projects and exchanging thoughts and over the course of the last year he actually quit his last career and he became a full time prompt engineer in Prague, Czech Republic. And he has been crafting these intricate prompts on an everyday basis for his company and the clients of his companies. And as Rastislav has been developing this, he has been consulting with me on a regular basis on how to lay this out so it's usable for you. And with the AI Advantage community we've been testing and iterating on this since while the result A brand new AI Advantage GPT named Sam the Prompt Creator. What does it do? It takes a problem you might have and it generates a complex prompt with all the context fleshed out. In other words, it's a really effective prompt generator. And we also built a function in where you can put in your existing prompts and it turns them into this more advanced version for you. That's this improve button. So I'm going to guide you through this. I'm going to show you how to use this here. And again this is mostly relevant if you're in the second bucket, if you're automating, delegating or if you're using a prompt a lot. If you just want to get something done in chatgpt you to want, you might not need this, but follow along as I show you this and make up your own mind. Before we get into functionality, just want to address the name. It's sort of a tribute to the head of OpenAI. Sam Altman is helping us get things done on a daily basis with ChatGPT and Sam the prompt creator is helping you to build prompts that just work. So how does this work? How does it generate prompts for you? Well, two ways really. One is the quick build. This is optimized to get your result as quickly as possible if I click this and a new version might look differently. We're iterating on this on a daily basis, but basically tells you it's here to build a prompt for you directly and you can just tell it what you want to do here. And all I need to do here is state what kind of prompt I want to build. So let's just take one of the classic prompt examples that a lot of people use. Email reply drafts. I want to draft email replies to my marketing agency building our website. And you can even go shorter if you want. Let me tell you, doing a proper sentence like this that gives it a little bit of details is helpful because then it will know more about what to build you if you want to make it as good as possible. That's the second mode that we'll look at in a second. But right here I'll just hit enter and there it goes. It just does its thing. It writes the entire prompt for you in a markdown format that you can copy with the click of a button and you're basically good to go and look. The point of this video is not breaking down all of the prompt engineering and the techniques that I used within here and within the GPT. And that's what we do in the community. We're holding a focus session. We're going to break down the build and the prompting behind this entire GPT in depth. All you have to know is that you can copy this, open a new chat and then run the prompt. Only thing that you might want to customize is the Free Shots template here in the end, where you can insert multiple examples of to how how you want that email to look. The product creator knows that it really helps if you do this. That's why it gives you this template. But you can feel free to delete this or insert examples here in between the lines. Everything else is set up for you and as mentioned, it fleshed out all the context. That might not be obvious at first sight, but if you're going to be doing this over and over again. If you have an automation that will be drafting email replies, you want to specify things like the task description in more detail with some extra context. What are the action steps that you're going to be taking? How are we going to be addressing issue resolution? If this occurs when you're working with agency, there's probably going to be a lot of back and forth and your ideas clashing with their implementation and your vision being different from their vision. We built this to understand a lot of these nuances of various use cases and includes the solutions to them into the prompt for you. You don't have to think about it. It specifies goals, outcome expectations and it adds a block with essential and important guidelines. And this is one example and it took a long time. You can go ahead and build prompts for yourself at no cost with this GPT we have and in the second mode it will help build your prompt with more in depth information. So if you have three to five minutes to answer some more detailed questions, this interactive questionnaire that we prepared in here for you will guide you with questions that help you zone in on what you really need from that prompt. So if I'm doing something for educators, I'll just say B and all I'll do is follow this interactive quiz and in the end it will generate the prompt that will solve my issue for me. In the second step I do the same thing I did with the Quick Build. I'm going to take a different route. Say I want to create interactive quizzes for my prompt engineering students and then as you can see, more multiple choice questions that become more and more focused on your use case. I will just fast forward to all the sections and then we can look at the result. There you go. After going through all the nine sections, it crafts this prompt for me that now is obviously way more dialed in than the Quick Build. And then last but definitely not least, it's the prompt improvement where it goes ahead and asks for the prompt you already have and it fleshes out the various pieces of context for for you. I'll take this organizational prompt that we featured in our recent newsletter, Paste it in here and again, in just one interaction it does its thing and improves the prompt for you. Now as you can see here, I use the prompt formula with square brackets where you need to change specific variables. I'm just going to share the prompt in here and I get the improved result. Two notes on this. Firstly and importantly, you want to use finished prompts. You don't want to use formulas. So if there's any undefined variables, square brackets, curly brackets, anything that is not final, this will not work consistently and you might run into troubles. So always use prompts that are specific where it does not include certain parts that are yet to be defined. Secondly, ChatGPT and especially GPTs have massive limitations. We did our best, but if something doesn't work as expected, simply run it again 8 out of 10 times things work as expected, but if you run into any sort of troubles and maybe you're getting unexpected results, just Go ahead and rerun the prompt and give it another chance. Ah, and good that I did that. Now we're getting an unexpected result. So instead of being confused by it, I'll just run it again. Actually, I'll add a little block here saying here's my prompt. Word of warning, we just released this, we'll be iterating on it, so if you run into any hiccups, just run it again. And we'll be implementing improvements over time. And that's really it. So if you're getting something done yourself, just quickly prompt it. But if you want more consistency, quality and detail, then you can use Sam, the prompt creator to either build a prompt quickly, build it more in depth by answering a series of questions, or you can simply upgrade the prompts you have and they'll work more consistently over time. And this goes for all models. OpenAianthropic, Google Meta doesn't matter. Giving it more context will improve output quality and increase consistency. So in conclusion, no, prompt engineering is not irrelevant. There's just situations where that might be the case. And a lot of the applications that are being built are going in a direction where they're trying to reduce friction with the users. They don't want you to type in custom prompts, and if they know what they're doing, they'll be doing something very similar to what I showed you here today. They'll be fleshing out the context so they get consistent results from the application that you're using. And the users are only going to see a nice button, just like Apple announced with their implementations. Right? You just say shortened text. You don't have to prompt it and tell it what short means. And you don't have to bother with fleshing out all the context around what a good short text looks like. They'll do that for you, but Sam, the prompt creator can also do that for you, but you retain a lot of the manual control. All the links are in the description. A big thank you and shout out to rastislavtuzava again for co creating this with me in the AI Advantage community. You can also find his LinkedIn below in the description. And with that being said, good luck out there using these large language models and I'll see you in the next one.