Introduction and exam overview

0:00 We're going to do the usual thing that we do, which is uh doing lectures, hands-on labs, uh and I have a free practice exam for you so you can ace that exam. If you like free courses like this one, the best way to support more courses is to purchase the additional uh study materials that will get you access to more practice exams, flashcards, cheat sheets, downloadable lecture slides, and more. Um, and I've taught a lot here.

0:23 So, I've taught a lot of Google courses, Adabus, Azure, um, Terraform, Kubernetes, and more. And basically all the AI courses in cloud. I'm the one who's teaching them. So, you're in good hands here. But uh let's jump into the course and I'll see you soon. Okay. [Music] All right. We're taking a look here at the Google Genai leader. And so this certification is a conceptbased Google Cloud certification teaching you fundamental geni concepts Google cloud's gen offering specifically uh for enterprise consideration as that's what they want you to know.

1:09 Um I need to point out that this is not a complete foundational course. Um there are other competitor courses like the AI 900 and whatever the one is called for AWS and I would say that they go in greater depth in terms of Genaii concepts. I have my own Genai uh essentials course which greatly covers that fundamentals. Um I would say that the amount of Genai concepts here is enough uh for who they're targeting for here.

1:35 But I would just say that it is incomplete and you should just be aware of that. Uh Google doesn't use course code. So we call this the GCP GAL. uh based on the G, the A and the L. So I'm calling the GCP gal as we go through here. This is a very easy certification. In fact, it is the shortest and easiest Google Cloud certification. So if you're looking for an easy win, uh you don't mind paying the whatever it is 99 USD and you want to show that you have those skills or just make sure you go through it, this is an easy one for you folks.

1:58 Um I would consider taking this if you are adopting Google cloud geni offering. If you're a key stakeholder, a solution architect or sales engineer and you need to understand the gen offering. Um but this is not a very much hands-on course. You can literally pass this with no hands-on experience. But I do make sure to show you things in the console um just so that you can contextualize it for your own use cases. Okay. I would recommend that you take the cloud digital leader first um because I'm going to jump right into the console and my expectation is you do have some knowledge in GCP first.

2:43 You could absolutely take this without taking the GCP digital leader first. But you know, you might need to spend a little bit more time on it. Again, very very easy course. I do need to remind folks the certifications not validate programming, technical diagramming, code management, many other technical skills. So, if you think that you're going to become a Genaii engineer after this course, you're going to be far from it because there's not enough hands-on to do it.

3:06 But for those in decision- making or need to know the offering, it'll be fine for that. How long to study? 5 hours. If you're beginner, again, I recommend that GCP CDL there and you can do it with no technical experience. If you are experienced, it's going to be a very easy course to complete, maybe two hours. I'd probably set this at 3 hours in the middle. Um, I'm saying 40% lectures and labs and 60% practice exams.

3:31 I rarely go this heavy. Usually I do this when we're talking about professional certifications. Um, but because there's not a lot of lecture lab content, uh, you do need to spend more time on the practice exams. Not that they're hard, but um, we'll talk about why in just a moment here. So, watch those lectures and labs. I would say you can watch the labs. You don't have to do them. In some cases, I will say like don't do this lab because I don't want you to run into spend or we're just exploring it.

3:53 So, understand that followalongs are not that necessary for the certification. Do paid practice uh practice exams for this one. You absolutely want to do them. And it's not because the questions are hard, but they have tricky wording. Okay? because sometimes when exams are too easy um they end up using very generic language like business language and then it just kind of muddies what it is what the answer is and so we try to reflect that the best that we can um and so that's why I recommend that you do practice exams okay you can take this via criterion through web accessor that is the platform that criterion uses for their online test centers if you can do it in person I strongly recommend doing so they do use AI proctors these days.

4:40 So a proctor is a supervisor person who monitors the student during the examination but they are leveraging AI um to monitor you. So just be aware of that. It's becoming common now. They've probably been doing it for longer at least. Criterion tells us other providers don't but they definitely are using AI. Um in terms of the breakdown here we have fundamentals of geni Google clouds geni offering improving the geni model output business strategies for successful geni solutions.

5:03 Each domain has its own waiting that determines how many questions in a domain that will show up. Um, passing grade 700, so 70% as per usual. If you get exactly 70%, you could fail. So, make sure you aim above that. So, when you do practice exams, I say go try to get 85%. Because usually people are off by 10% and so 75% gives you that buffer in terms of passing. There's 50 to 60 questions. I got exactly 50.

5:26 I didn't get even 60 questions. You can get about 18 wrong here. There's no penalty. The format is multiple choice. I don't think it was multiple answer. I think it was just multiple choice. Um, but anyway, if it is, it's not like it's going to be that hard. You only get 1.5 hours. So, normally they give you two hours, but they only give you 1.5 on this. So, I say about two hour 2 minutes per question, but it's probably less than that.

5:49 But again, you'll finish this very, very quickly. Um, the exam time 90 minutes, it's 1.5 hours. buffer for 30 minutes. So, you can um review the instructions, show the online proctor, your workspace, read and accept the NDA, complete the exam. Criterion I don't find very stressful. Um my laptop's broken right now and I have a a very strange setup and they're very nice and accommodating and uh despite my very uh strange looking setup, so I was able to sit this exam no problem.

6:12 I wouldn't stress out about it. Though, I do find with Criterion that um sometimes the phone check-in doesn't work for whatever reason. their their their mobile web app is not very good and so I just try to force it to do the desktop check-in. Um but yeah, the secure the secure web accessor can be kind of a pain. So make sure that you show up early for this so you can avoid that issue there. Uh this is valid for 36 months, so 3 years.

6:36 Um and there you go. So yeah, it should be an easy pass for you. Just make sure you use those practice exams to fill any gaps or tricky wording and you'll be fine. Okay, chow chow. [Music] Hey folks, it's Andrew Brown. Let's go through the official exam guide so I can give you some thoughts and opinions to prep you for this certification. I did sit this exam of course and actually I sat it blind and I passed.

7:22 So that means I went in and I didn't study anything and I passed and that's because I have knowledge from other providers. I'm already familiar with Google's offering. I'm going to tell you it was a really easy exam. uh some of the wording was a bit tricky. Um it's just again because of the way they use business language and describe things. So some things could have been uh worded better but for the most part um if you go through the materials you should be able to pass and definitely do the practice exams to deal with those tricky questions.

7:45 Um so section one is fundamentals of geni. I said earlier that I do feel that it's incomplete but it's sufficient enough. So they want you to know AI, natural natural language processing, machine learning, geni, foundational models, multimodal foundation models. They might have a question like what is AI, right? And they might have a question like what is NLP? But the rest here didn't really get a whole lot of questions. Um I packed in some of my foundational content from my Genai Essentials course.

8:09 Um so you know, I could have made very specific uh content for that, but I don't feel that it was necessary. Okay, so I just brought that over and that's frontloaded in the course just the materials that you need. Um, you know, so if you hear mention of other providers, that's why because I brought that over. Um, but any we'll continue on here. So diffusion models, I don't think I covered that because there's no questions on the exam, but diffusion models is the type of models that we're utilizing.

8:32 I do have this in the gen essentials, but it becomes very technical and that was out of scope for that, so I didn't put it in here. Prompt tuning is knowing about the parameters, temperature, top, P, things like that. prompt engineering. We cover the types of prompt engineering in here that that they will ask you. They will actually ask you like you know what is a oneshot, what is a zero shot chain chain of thought. So we have lecture slides on that.

8:55 Very simple stuff. Um describing machine learning approaches. So you need to know supervised and supervised and reinforcement. So I have some slides on that for you. Identify the stages of machine learning. Um this didn't really come up on the exam. Um I do have like a generic one in the Google cloud digital leader which might have been brought over but there was no questions about this. So I don't feel that this is necessary in terms of passing the exam but you should know it because it's good to know.

9:20 See identifying how to choose the appropriate foundational model. Um there wasn't really any specific questions on this. It was more like like um if you had I'm going to get my head a little bit out of the way because I know I'm covering some of the text here. But coming back to this one here, it was more like, oh, you're using Google Gemini or like you need to use a multimodal um thing and you have to deploy it in your own your own um you know local compute.

9:45 So you can't use Google Gemini, but you can use Gemma. And so it would be as simple as that. So if you know generally what the models are, it's not going to be that hard to figure out what to use. Identifying business use cases for geni, creating summary, uh discovery, text generation. So if you have a general idea of the functionality of Vertex Studio AI, you're going to know what to use. I'm describing how various data types are used in Genai.

10:09 There wasn't really anything about data types. So there was nothing really to make on that, but we do cover data types in the course. So if there is something there, it'll get covered. Explain the characteristic importance of data quality, data accessibility. No questions whatsoever on this. So I don't know why this is in here. Um nothing nothing whatsoever. I didn't get any. I sat it and also my uh my um support engineer Peter sat this exam.

10:31 We both didn't get it. I don't know why that is in there in the exam. Identifying the difference between structured and unstructured data. Identifying real world examples of each type. You don't really need real world, but you do need to know the difference between structured and unstructured data. So that's pretty straightforward. Labeled and unlabeled data. If you know what labeled data is, it's data that's been manually labeled by a person. Then you know what unlabeled data is.

10:53 It has no identification. uh describe how various data types are used in Genai. Um so again I mean like this just comes back to like structured and unstructured data and labeled unlabelled data. So maybe that's what they mean by data types but like data types is a generally mean something even more specific which wasn't really covered. So they must mean just unstructured and structured and and labeled and unlabeled and we cover that thoroughly in the course.

11:16 identify the core layers of the gen landscape infrastructure models platform agents applications. There is no questions whatsoever on this. So, and I mean this is pretty straightforward. You know, the infrastructure is what runs it. The models is what serves up the AI stuff. The platform is the the comprehensive integrations of all their toolings. Agents is um a model that uh like acts like a human and then you have your application on top of it. So, there's really nothing to describe here in the course, but um you can see the layers there and if you had to identify them, I'm sure you'd figure it out.

12:03 It's not too complicated. Identify the use case and strengths of cloud uh Google's foundation models. So here they want you to know the difference between Gemini, Gemma, image and v video v video v video v video v video v video v video v video v video v video vo I'm going to tell you right now Gemini multimodal it's fully managed has different integration or extra integrations with API Gemma is just a model it's open weight you can download on your own uh your own machine um some of them are not multimodel the older ones aren't but most of them are images for generating images for generating videos if you remember that you'll get those pretty easily but we do cover them so you'll be sure to understand them um so let's take a look here so we Google Cloud's Geni offering.

12:27 So, describe Google's Geni first approach and commitment to future innovation. This is just like marketing nonsense. So, if you get questions, you just have to best guess on the marketing questions because there's nothing in particular that I can point to. Um maybe within um Google's official course, they might have some some kind of text here, but I don't think you'll lose any points here or you have to go seek out that other content. You'll just figure it out when you do the exam.

12:51 describe how Google enterprise ready AI platform is a responsible secure private reliable scalable I don't know like they they have again generic generic onepage marketing things and you just kind of guess um and it's like the most obvious answer answer and you'll get it get it so that's totally fine recognizing advant advanced Google's comprehensive AI ecosystem um I mean this is no different than any other provider so just think of like why it would be useful to have everything like it's you don't even think of geni just think of cloud and you'll you'll figure it out describes benefits of Google cloud's open approach I think I had one question about this where was like people were concerned about like vendor lock in and so um you just had to choose the question that's that was the least uh concerning so like if you were adopting it just think of like what answer sounds the best to you for

14:00 adoption and you'll get that answer but there's nothing like in particular again um in the marketing pages about open approach maybe they just mean like Gemma has open weights and you have some options of other open models. Um, but other than that, I don't really know what they're talking about. Identifying the essential components of Google Cloud AI infrastructure. So, they might ask you questions about TPUs and GPUs, but I didn't get anything about hypercomputers. Uh, if you know what the the the TPU is, um, then you'll be fine.

14:21 So, we have a slide on that. I don't even think they updated it. There probably is a new TPU, but they're not going to ask you like what kind of TPU. Um, it's a tensor processing unit. You just need to know that um Google makes their own um tensor processing unit specifically for machine learning um and GPUs can be used but these are more efficient for working on tensors. Um and they don't even talk about the tensor tensor uh tensorflow framework which is why use tensors.

14:44 So I'm surprised they don't have that in here but that's fine. Explain how Google Cloud AI platform provides user controls over their data. Uh yeah, so there was questions about um uh that kind of stuff, but that might be like about like service agents and things like that. If you've done the digital leader, you can just answer those. There isn't anything in particular that was specialized for it, so there's not much to say. Describing how Google platform is democratized AI.

15:07 Um again, nonsense question. There's no real questions on that or just guess your best for that. Describe how Google's pre-built jetting offering enables AI powered work. So recognizing the functionality, use case, business value of Gemini app and Gemini advance. So we cover what gems are. Gems are just the same thing as um chat GPTs GPTs. It's just a template for if so you don't have to like fill it in every single time. And then like we look at Gemini, they say Gemini app, but I guess they just mean the generic Gemini application.

15:30 I didn't get any questions actually about the Gemini app. But like if you know what chatbt is, there's the Gemini app. Like it's straightforward. They do have this thing called agent space. So they're probably that's what they're they're talking about because agent space is like the Gemini app uh enterprise offering. Yeah, they mentioned it right here next, right? So that's probably what they're talking about. Um cloud notebook LM. So we look at that. They say API, but you really don't need to know the API.

15:53 You just need to know what notebook LM is. Multimodal search custom agent capabilities. I think we cover those. Again, there's no questions on those. You just need to know what notebook LM is and Google agent space. Recognizing functionality use case specific values of Gemini for Google Workspace. Um this is just you have Google uh Gemini within Google Workspace. That's it. That's all you need to know and then it can work in your docs and your your email and whatever programs they have.

16:37 It will work in there. Describe how Google Cloud's geni offering improved customer experience. Um so again it's just your best guess. Um well not necessarily here because it's talking about vector AI search and Google search. So if you know um vector AI search is a out of the box rag and has specific integrations you'll be good and the fact that you can do grounding with Google search and with vector AI search then that'll be straightforward as well.

17:00 Um here we have the customer engagement suite. I made sure to thoroughly cover this. I got a lot of questions on this um and um so you know I made the videos for that and I do the labs for it. You don't have to do the labs but again I recommend watching them so you know fully how they work. describe how Google empowers developers with AI. Um, so the Vertex AI platform. So, model garden, vertex AI search, AutoML, AutoML never came up once whatsoever in the exam.

17:22 Um, I don't even know why this is in here because I didn't get it. Peter didn't get it. I didn't make a slide on it, but I do cover in the Google digital leader. AutoML just means um it's like a machine learning model that is uh it deploys all the stages for you and you can just choose some parameters but it's in the digital leader but I don't think we put it in this course and you're not going to get a question on it.

17:45 Um and we wouldn't even know what to write a practice exam question on because we didn't see it. U model garden is a model catalog. You'll see that so many times it'll be very easy to understand what it is. Recognizing Google cloud rag offering. So they hear they're reinforcing about the Vert.ex AI search about um the built-in stuff for that. So that's pretty straightforward. Um define the purpose and type. So identify how agents and tools interact with external environments to achieve their tasks.

18:10 We could be talking about rag here um or tool use or things like that. Identifying relevant Google cloud services and pre-built AI agents for agent tooling. So, uh, Asian tooling again is just calling out two things. Um, and so they're just saying there's a variety of things that you have here. I didn't really get any questions like this like Google video intelligence. I didn't get that. Document AI API, didn't get that. Document, like none of this stuff, but this stuff is pretty straightforward.

18:33 Like, you know, they have text to speech API if you want to use it, right? Um, it's pretty straightforward. And some of these things actually get covered in Vert.exai. So, Vert.exi XUI like puts a layer on top of this stuff. But I guess here they're saying like these are um things you could specifically use for tool use. But again, there was no questions on a lot of these things here. So, but again, it's very straightforward what they are.

18:57 Describing when to use Vertex AI Studio versus Google AI Studio. Um I couldn't tell you like I've used both of them. Google AI Studio is is um if you don't have a Google account basically like a Google Cloud account. So if you don't have a Google cloud count, you probably want to use Google AI studio. If you want more model access and more integration with Google cloud, you're going to use Vert.x AI studio. That's it, right?

19:21 Um if you use them, it's very straightforward like what they can and can't do. Techniques to improve geni models. So identify common limitations of foundational models like data dependency, knowledge cut off, bias, fairness, hallucinations, edge cages. So these were I got questions for this, but I put these in the practice exams. So, I don't think I made slides for this, but um I definitely made practice exams uh sorry, practice exam questions for this. And so, you know, if you had a question about data dependency, you know, so again, I'm just going to tell you right now, make sure you do the free practice exam so you get the coverage for this one because literally there'll be like one, two, three, four, five, six.

20:07 You'll literally get six questions in your exam. And so, if you can get these in the practice exams and experience that, you're absolutely going to get six points in your exam. No problem there. describing the Google Cloud recommended practices to address limitations. I don't know where their pra uh their best practices are. They're not in the docs. Um so whatever I did is what I would recommend. So whatever I do in the labs, that's going to be that.

20:33 And again, remember I passed this blind. So you know if I can pass it blind, I think you'll be okay. Recognizing Google recommended practices for continuous monitoring evaluation gen. So automatic upgrade models, key performance indicators. I think I had one question about KPI, security, patch, updates, versioning, performance, tracking, drifting, monitoring, vertex AI, feature store. A lot of these weren't on the exam to be honest. Um, but you know, they have them here in the the exam guide.

20:55 Define prompt engineering. Describing it significance in interacting with large language models. Identify prompting techniques and use zero shot, one shot, few shot, role, prompt chain. Definitely had a question for each of these. Okay, so I covered those in lecture slides. I think I show you how to do those and the practice exams at least have one of each of those. So that again is like one two three four five points. You see like if you do this five points six points you're basically halfway there.

21:19 You're getting there. Identify advanced prompting techniques when to use. Um we do have react prompting but they didn't have any questions at least I didn't have. I did have a question on chain of thought prompting. So that's great. Describe the concept of grounding. So they're coming back to another section and repeating uh vertex AI. And so grounding is basically going out and getting data from a rag um and making sure your data is up to date.

21:43 And they're talking about rag again and that stuff there. And they're talking about pre-built stuff. So literally they repeat this like three four times. So you can see how this is a padded out u exam guide. Then we talk about token count, temperature, top p, safety settings, output length. So I made sure to include a a lecture slide on that. I always show it in the labs but I went out of my way and made a a lecture slide for that this time around.

22:06 Um, describe the Google cloud recommended steps to successfully implement a transform and genai solution. So here we have text generation image. So this is just going through and knowing all the options of vertex AI studio. Uh, again this is just guessing what is the right thing. Describing how to use the right geni solution. Again this is just guess questions. So this is just fluff in here. describe security and the importance in protecting AI systems. So there is no questions on the SIF.

22:28 It's like a single page document and it's complete marketing nonsense. Um but you know I cover it. Same thing with this one as well. Secure by design gets mentioned. Um so if you see a question that says that it's an option for secure by design just choose secure by design because like again I couldn't find much information on it. So there was that. I remember I got a question on security command center. So if you know what that is, that's important.

22:52 I am they're just talking about service accounts. So you know whether you should grant service accounts or not to something. If if we have a lack of this content in video form, they're definitely in the practice exams because I'm telling you, we wrote every possible practice exam question and make sure you use the practice exams. I even tell you in the intro 60% practice exam. So I'm just going to emphasize that over and over again. Um, if there's something not in here, it's I felt that this should just be in the practice exams and you can just learn through that way.

23:15 Explain the importance of responsible AI and transparency. I revisited the responsible AI page. Now, they used to like if you go to Adabus in Azure, there's stuff that's more in detail, right? So, I went responsible AI Google and like I there was just nothing here. They go here, you go, okay, it's like all right, where is it? Then you read about the principles, right? and they like bold innovation that is not helping us here whatso like what is that so this is what I mean when I say there's a lot of marketing nonsense uh that is in here but again this is aimed at people for key stakeholders uh sales engineers and stuff like that so maybe they want you to know that marketing that marketing stuff to carry forward to your customers but I think it's a bit misaligned and to be honest you can just guess best for that and you'll be fine um explain the

24:27 importance of no we already read that describe privacy considerations so risk privacy data anomalization so this stuff you'd want to know um usually use DLP for that data loss protector so DLP GCP so if you want to mass data or you detect it or whatever like obscure it deidentify it mask it tokenize it it's going to be DLP okay so that's your answer right there DLP Okay. Uh I don't know if you made a slide on that.

24:51 Just DLP. Remember DLP describing the implications of data quality, bias, and fairness. Um again, this stuff is more covered in the um the practice exams. I didn't make slides on those and this is just a repeat of what was up here just a moment ago, but you know, you want to know them all because you'll get six points. Okay. Describe the importance of accountability, explanability, and AI systems. Um, and so explanability, I don't think I made a slide on this either.

25:15 Explanability AI, um, Google, we'll just take a quick look here because there's no like when you use like Azure and AWS, they actually have very prescriptive tools that do explainable AI. Like they're literally specialized models. But for Google, it's just Vertex AI, explainable AI. And Vert.Ex Ex AI explainable AI just explains how the model works, right? So if you get a question where there's uh it's like a customer's cons like is you know um it's an insurance company and um you know it's making biased claims uh against a minority group um and the company needs to prove that they're not being unfair.

26:06 Explainable AI will explain how the model made its decisions and that's all you need to know to answer that question. Uh, so yeah, there you go. I mean, again, it's not a hard exam. I'm just going to point out explainable AI, right? Security command center. That was another one there. DLP over here. Um, and try to hit like uh come back up here for a second. Like know all these. They're in the practice exams. This is also as lecture slides.

26:32 And the other one was these here. These are only in the practice exams. I don't think I made slides on this because it'd be really boring to make. But anyway, there you go. and have fun and see you soon. Okay, cha chia. [Music] >> Hey, this is Andrew Brown and we are