AI and Prompt Engineering

Oracle Academy Tech Chat

In this episode host Orlaith Lawton speaks with Oracle Cloud Engineering expert Alex Negu about AI and Prompt Engineering. --------------------------------------------------------- Episode Transcript: 00;00;00;00 - 00;00;35;19 Welcome to the Oracle Academy Tech Chat. This podcast provides educators and students in-depth discussions with thought leaders around computer science, cloud technologies and software design to help students on their journey to becoming industry ready technology leaders. Of the Future. Let's get started. Hello and welcome to the Oracle Academy Tech Chat, where we discuss how Oracle Academy prepares the next generation's workforce. 00;00;35;21 - 00;01;13;29 I'm your host, Orlaith Lawton. And today we'll be talking about AI, a prompt engineering in the cloud. In this episode, I'm joined by my guest, Alex Nagel, an Oracle expert in cloud engineering. Alex is a senior cloud engineering manager at Oracle. He's responsible for overseeing projects in Central and Eastern Europe. Since joining Oracle, Alex and his team have tackled some of the most challenging cloud projects helping customers leverage the innovative services that Oracle offers from start-ups to large enterprises and even the public sector. 00;01;14;01 - 00;01;43;04 Alex has led over 100 cloud projects. Recently, he has focused on ground breaking achievements in artificial intelligence with expertise in areas ranging from data management to large language models. Alex is dedicated to facilitating A.I. adoption across diverse industries. To start off, can you give me a little bit about your background, Alex, and what you do at Oracle? Absolutely. 00;01;43;11 - 00;02;03;24 My main job as a cloud engineer is to help our customers get the most out of their cloud services. Basically, we work very closely with them, advising on everything from the design of the systems to making sure everything runs smoothly. I lead an awesome team, by the way, which means I get to learn from each of them and from different projects being involved in across central and eastern Europe. 00;02;03;27 - 00;02;21;25 What's really cool about this role is that no two days are the same. We work with companies from all sorts of industries and each one comes with its own unique challenges. This means we spend a lot of time researching and finding the best solutions. It's a great way to keep learning and growing. Probably that's my favorite part of the job. 00;02;22;01 - 00;02;42;19 Being involved in these projects, just to give you some examples, has given me insight into how financial system works, how airlines operate, how farmers take care of their crops, and how people in end use are tackling the most important challenges the one that seems to be unthinkable like plastic pollution in the Mediterranean Sea. You know, that's something really inspiring. 00;02;42;21 - 00;03;00;26 Something I always like to highlight is the bigger picture when you're thinking about cloud and technology in general. Remember that those are just tools to solve the problems. You start with a challenge and then figure out what pieces you need to solve it. It literally is like we are playing with puzzles; we're trying to solve riddles every day. 00;03;00;28 - 00;03;22;00 I think I have one of the coolest jobs in the world. You know, our role getting back to the initial question is to make sure we are solving our customer issues in the best and most optimal way, whatever that means. Yeah. We get to work with all kinds of exciting stuff, things like local development, blockchain automation across the stack, and of course artificial intelligence. 00;03;22;00 - 00;03;46;27 But at the end of the day, it's all about solving real world problems. Wow, that's a huge amount of information. It sounds extremely interesting and exciting, but great that you're sharing that with us. You mentioned A.I. Alex, and it's hard not to notice all the buzz about it in recent years. Can you share maybe your thoughts on where A.I. fits into the bigger picture? 00;03;47;00 - 00;04;08;20 AI has indeed been a huge topic lately, especially with the rise of consumer-friendly tools like J.G. Beattie. But really, it's important to separate the hype from reality. Right. I mean, it’s everywhere, but I think we need to get back to the basics. First off, we need to understand what AI is. At its core is about making predictions. 00;04;08;21 - 00;04;32;00 Yes, it's incredibly complex, but ultimately, it's still about making predictions. There's a lot of buzz around AA right now, and some people think we'll soon have artificial general intelligence, the kind that can perform cognitive tasks better than humans. However, I'm talking from my own experience. The closer you look, the more you realize that's not quite the case. We are not there yet. 00;04;32;02 - 00;04;56;07 He's been around for decades. It's not something new. What's new is the breakthrough ingenuity VR, which is just one part of a much larger and broader field. We've been using it for things like computer vision, gesture recognition and predictive maintenance for a while now. This use case has been across various industries, right? So it's not specific to computer science or something very linked with the tech only. 00;04;56;10 - 00;05;19;13 The big shift recently is how accessible it has become for everyday consumers, just like myself or yourself, when we are not working here at Oracle, Tools like Jeep DS Make it easy for anyone to use AI, and because these tools mimic human like intelligence, they've seen a huge surge in adoption. As for where AI fits in the bigger picture, I believe it will blend seamlessly into our daily lives and activities. 00;05;19;18 - 00;05;44;18 So without significant new breakthrough breakthroughs, I don't see AI replacing humans. Instead, it will augment what we do. Making books easier and enhancing productivity over time will become a commodity, something we take for granted, just like many technologies before trade. Don't get me wrong, AA holds immense potential for addressing social issues such as poverty, inequality and access to education. 00;05;44;20 - 00;06;09;19 Areas where humanity hasn't done enough so far. But for individuals, it can boost productivity and simplify our lives. That's why I think it's crucial for everyone, especially students, considering the toxic context in which we are meeting to keep an eye on their development, Stay updated with the latest advancements. Wow. That's really interesting. And tell me, Alex, you know, as a student, listen to this. 00;06;09;21 - 00;06;33;15 You're probably thinking, do you need a computer science or a master's degree perhaps, to get started with a I? No, no, I don't think You don't. I don't think you need it. But look, it might depend. Yeah. It's important to distinguish how each of us interact with there. How do we position ourselves into the context? For instance, you use the internet without needing to be a network engineer. 00;06;33;20 - 00;06;53;17 Right. Similarly, you can take advantage of artificial intelligence without having to be a data scientist or researcher or an engineer. Eight comes in many forms. It could be a featured in your favorite phone app. We all see that every day, right? With the new updates being rolled out. Or it can be a chat tool that you can use for various tasks. 00;06;53;19 - 00;07;18;05 Let's talk about the latter, since it's probably one of the most popular examples that everyone is looking into. More recently, tragedy, for instance, I'm sure everyone has heard about by now. It really has been a game changer over the past few years. It offers a chat interface powered by a highly capable large language model. You can use it for searching information, creative writing, proofreading, brainstorming, problem solving, and so much more. 00;07;18;08 - 00;07;46;06 The beauty of it is that you interact with it through natural language, making it the bot intuitive and accessible. And with the addition of new multimodal capabilities, like how the new models are being released, you can now interact using not only language but voice or images as well. Expanding the range of possible use cases exponentially. Looking ahead, there will be people driving innovation and those will be those who will utilize it. 00;07;46;09 - 00;08;13;15 I'm going back to your question. No, you don't need to be a specialist to achieve great things. Be there. You just need to learn how to use it effectively. Excellent. Okay. And you mentioned the large language models. So let's imagine it's my day one when it comes to using large language models or EVs. What should be the starting point in learning how to use these elms before you dive into using a lens or a model? 00;08;13;17 - 00;08;34;18 Right. It's important to understand how they work, what they're good at, and especially their limitations. You don't start by the car without knowing where the steering wheel is and how to press the brake right. So knowing the basics will help clear up many questions you might have later. Keep in mind, these models might not have access to specific info you're looking for. 00;08;34;24 - 00;08;52;08 They can also get the things wrong or make stuff up. You know what is called hallucination? So easy to spot. They create answers from the ones which are flawed. You don't have to be a pro, but you should know the basics. I mean, you should have a basic understanding after you get the hang of how this model works. 00;08;52;10 - 00;09;19;07 You can focus on prompt engineering. So what is a prompt? It's basically the instruction you give to the AA, the message you're sending. The output is based on what you're told, right? To get the best results, you need to know what was and how to ask it. Prompt engineering is all about writing clear and precise instructions to look closer to a good prompt usually has three parts the context action and the guidelines. 00;09;19;14 - 00;09;41;23 So context. This is if you want the big don't know the background info that the model needs to answer your question. For example, if you're asking questions about yourself or your family, the model needs that info to respond accurately. Without it, the model might give a wrong answer because it was not trained on personal data, so it doesn't know things about the question you are asking. 00;09;41;26 - 00;10;00;03 Then moving from context, you have the action. This is what you want the model to do. Are you asking me to write the poem? Give suggestions? I don’t know. Brainstorm some ideas. You need to be very clear and create about what you are asking for. This is what the action and the instruction you are passing is all about and the guidelines. 00;10;00;06 - 00;10;22;27 These are all the extra details like how you want to answer to how the answers should be formatted, the tone and the role they should play. Right. These are small details, but it makes a huge difference when it comes to the total response you are going to get. So that's the basic idea. Some tasks, for example, might be so complex that you need to break them down into smaller steps. 00;10;22;29 - 00;10;41;20 Also, you might need to give examples or explain exactly to the logic which model what you want. There are plenty of things you can take and this is a field, you know, it's under severe research. You need to keep an eye on what they are. In this one, it's more on what kind of improvement you get in that area. 00;10;41;23 - 00;11;03;08 Trust me, this is a really exciting field and once you start to want, we will not get bored. It sounds really exciting. It's really interesting, actually, the terms that you're using, the hallucination, for example, that's an interesting concept that I hadn't heard of before, but maybe we might chat about that a little bit later. But getting back to the models, which is the best model that I should be using. 00;11;03;10 - 00;11;27;14 Yeah, that's another great question, but it's not as straightforward, straightforward as it might seem. Let me ask the question in response. Right. What is the best sports team in the world? Right. I hope you get my point. There isn't a one size fits all answer here. The best model depends on the specific task we have in mind. Models may share underlying principles, but they differ in terms of the dataset. 00;11;27;14 - 00;11;45;10 They were trained on the size, the intended purpose cause speeds and many other things. The largest model, again, to take a look at an example, the largest model and the most capable one in the absence of the internet is useless, isn't it, since it cannot run on your phone. You have a large model, but you cannot access it. 00;11;45;15 - 00;12;04;14 But the smaller model might work even when disconnected. So it can run on your phone. It's going to do a much better job since it's accessible and you have it there, right? There are plenty of benchmarks and methodologies and discussions around this topic how to assess the model performance. But it's important to approach them with some skepticism and rely on your own experience and judgment. 00;12;04;15 - 00;12;28;12 That's the only advice I can I can I can give here. For instance, some models excel at general knowledge, while others might be better at reasoning, content retrieval or specific niche tasks. My advice to everyone is to dive in and experiment is the most intensive research the thing right now, but nobody holds all the answers since it's relatively easy to access various models. 00;12;28;19 - 00;12;49;12 You can go and try out a few of them. Each model has its own quirks and strengths, so hands on experience will be your best. Gotcha. Okay, so experiment and stay curious. Curious with it. Can I throw in an extra little question? I just had to see it. And would you mind maybe just describing that a little bit more detail? 00;12;49;14 - 00;13;14;09 Maybe Give us an example what an observation is, because that sounds quite interesting. I haven't heard about that large language model thread. Those are programs in the end, right. And running on a computer like actually it's a very large cluster of computers that are working together to do intensive mathematical calculations. But those computers have a task and the task is to provide an answer, and they will provide that answer even if the answer is not accurate or not. 00;13;14;09 - 00;13;37;25 Right. So let's imagine you're asking a question that the large English model doesn't have. How I mean, it was never trained on, right? So you're asking a question about something that's something that has happened after the date at which the model was trained. Yeah. So the model was trained last week. And you're asking a question that about something that happened this week. 00;13;37;28 - 00;14;00;06 Okay. So you will notice that most likely, despite the fact that doesn't have the fact and it was never trained on such a piece of information, the model will try to give you an answer. Yeah, or this is just a very basic example. Of course, there are these kind of guardrails. I mean, the producers of the English models are putting some guardrails and they're trying to limit this kind of behavior. 00;14;00;08 - 00;14;21;04 But this is an example of hallucination, right? So, yes, you can you can rely on the results and the feedback provided by the large language model, but always take it with a grain of salt. And I think it's you know, it's good to understand where the answer is coming from. So go back at the sources, then double check that fact, check what the model is providing. 00;14;21;07 - 00;14;42;19 Otherwise, there might be cases where hallucinations are going to be so convincing that you take it for granted. And it's not the path you want to take. That's really, really interesting. Okay. Well, then one final question. If you could give one piece of advice to faculty or to students, what would that be? that's that's easy. Keep learning. 00;14;42;22 - 00;15;03;29 Whether it's with or without the. So choose your own path, because there are many ways to be part of this revolution. You cannot you know, you cannot avoid it. You will not be left out. So not everyone needs to be an engineer, right? Find a professional field that you genuinely enjoy and don't make decisions solely based on market trends or hype. 00;15;03;29 - 00;15;23;06 This year is in five years. Might be something else will likely touch every corner of every industry, but it won't take away, you know, the real things, the joy and the satisfaction of doing what you love the most. So the last thing I would like to say on the topic, and that's also because it's my mantra as well, my mantra as well. 00;15;23;08 - 00;15;47;03 Stay curious. And that's the best way to stay ahead in any field. That's wonderful. And thank you so much. Just to summarize maybe a few points of what you're saying, stay curious, which is so important. I will definitely become part of everyday life and take over money corners of every industry, but find something that you enjoy. I think that's great advice to all of our students. 00;15;47;06 - 00;16;06;08 And in a way, that's how you can stay curious if you keep going with what you enjoy as well. So once again, Alex, I'd just like to say thank you so much for helping us in our academy. Spreads this information to our faculty and students. I know you've done so many times for us in the local office in Romania. 00;16;06;09 - 00;16;25;08 We much appreciate that. So it's great that we now have a global audience to share your expertise with. So thank you so much, Alex. Thank you. That wraps up this episode. Thanks for listening. And stay tuned for the next Oracle Academy Tech Chat podcast.