0:04 Hello everyone, my name is Dya and today
0:07 I'm here to talk about Google Cloud
0:11 Generative AI leader certification exam.
0:14 I recently earned the certification and
0:18 I'm here to share what helped and uh
0:21 some study materials. So let's jump
0:25 right in. In this exam there are five
0:27 modules that you need to study. The
0:30 first module is about the generative AI
0:35 uh beyond the chatbot. Then the second
0:37 is about unlocking the foundational
0:41 concepts and Genai navigate the
0:44 landscape. Genai apps transform your
0:47 work and Genai agents transform your organization.
0:50 organization.
0:54 So I have also linked uh some of the uh
0:58 study guide and how to register and even
1:03 some sample uh questions uh how you can
1:06 prepare for the exam that will be helpful.
1:09 helpful.
1:14 So let's start from the study guide.
1:18 The first topic is about introduction
1:20 uh to the genai leader
1:23 and it the exam focuses your knowledge
1:27 in four key areas. It's the found uh uh
1:29 fundamentals of generative AI which is
1:32 the 30% of the exam and then Google
1:36 cloud geni offerings 35% of the exam
1:39 techniques to improve genai model output
1:42 20% of the exam and business strategies
1:44 for successful geni solution it's about
1:48 15% of the exam and uh please note that
1:50 this study guide is meant to be a
1:52 starting point and the exhaustive list
1:55 of resources I have linked the resources
2:00 U of each topic in this uh
2:03 video and you will be able to take that
2:07 in detail but let's go in the uh
2:09 fundamentals of generative AI. So you
2:12 will see um these layers it looks like
2:15 the layers of onion. So what is the
2:18 outside layer? It's uh artificial
2:20 intelligence. What is artificial
2:22 intelligence? is the building machine
2:24 that can perform tasks that typically
2:26 require human intelligence such as
2:28 learning and problem solving and
2:32 decision. Within that layer of AI, you
2:34 will find machine learn. It is a sub
2:36 field of AI where machines learn from
2:40 data to perform a specific task. Then
2:42 within that machine learning, you will
2:45 find genai which is an application of
2:46 machine learning that focuses on
2:49 creating new content. uh the content can
2:54 be images, videos. So um coding is an
3:00 example of u u generative AI. So we can
3:03 use that to generate new content. And
3:05 what is a deep learning? Deep learning
3:07 is a subset of machine learning that
3:10 uses artificial neural networks with
3:12 many layers to abstract complex patterns
3:16 from data. And then um we will talk
3:18 about the foundation models. It's
3:19 powerful machine learning model trained
3:23 on massive amount of data um allowing
3:25 them to develop a broad understanding of
3:28 the world and a large language model
3:30 it's a foundation model that is designed
3:32 to understand and generate human language
3:34 language
3:38 uh so what is uh genai it can create uh
3:41 new content and ideas uh genai
3:43 applications can be multimodal enabling
3:45 them to process and generate different
3:48 type of data like text images, code
3:50 simultaneously and genai can be used to
3:53 create new content, summarize, discover
3:57 and automate the task. Then uh let's
4:00 jump in the foundation model. Uh large
4:02 uh AI models are trained on massive data
4:05 sets allowing them to be adapted to many
4:09 tasks. Uh they are basis of genai. What
4:11 are the foundation model features? They
4:14 are trained on diverse data. They're
4:17 flexible to wide range of use cases.
4:20 Adaptable to specify
4:23 specialized domains through addition
4:26 targeted training. Um we will now let's
4:28 talk about prompting. Prompting is a
4:30 method of interacting with the
4:33 foundation model and guiding them. It
4:35 involves providing them with
4:37 instructions or inputs to generate the
4:40 desired output. And prompt engineering
4:43 is an art um and science of creating
4:47 effective input known as prompts uh for
4:49 generative AI model to maximize their
4:52 value and tailor the responses to
4:54 specific need. And now uh we'll talk
4:57 about two types of uh data label data
5:01 and unlabelled data. Label data is like
5:04 uh that has associated tags such as
5:07 name, type, number and unlabelled data
5:10 it's kind of law raw unprocessed
5:12 information that hasn't been tagged and
5:15 it lacks meaning by itself such as
5:18 unorganized photos or streams of audio
5:21 recordings. Machine learning has uh
5:22 three primary learning approaches.
5:25 Supervised learning, unsupervised
5:28 learning and reinforcement learning.
5:30 Supervised learning, it trains model on
5:33 labelled data to predict the output for
5:35 new input. Unsupervised learning uses
5:37 unlabelled data to find natural
5:39 groupings and patterns. And
5:41 reinforcement learning learns through
5:44 interaction and feedback to maximize
5:48 reward and uh minimize penalty.
5:50 Um now let's talk about the fundamentals
5:54 of generative AI. Uh data is in is
5:56 information that can come in many forms.
5:58 number, date, text, description, even
6:01 images and sounds. So structured data is
6:04 organized and easy to search, often
6:06 stored in relational databases.
6:10 Unstructured data that lacks predefined
6:12 structure and require sophisticated
6:15 analysis uh techniques. Quality data
6:18 data is that is accurate, complete and
6:20 consistent and relevant. Accessible data
6:23 data for model training needs to be
6:26 readily available usable in proper
6:30 format. Genai landscape uh geni powered
6:33 application uh it's a userfacing part of
6:35 generative AI. This is the layer that
6:38 allows users to interact and leverage
6:42 the capabilities of AI. Then what is an
6:44 agent? An agent is a piece of software
6:48 that learns how to best achieve a goal
6:50 based on input and tools available to
6:53 it. So the tool is important here. It's
6:57 a distinguishing um term when you come
7:00 um to agent. uh there will be many
7:03 questions about it and I think in my
7:06 exam there were at least uh two to three
7:09 questions about agent and uh the
7:13 differentiating factor was the tool
7:17 uh platform it's a layer offers APIs
7:20 data management capabilities and model
7:23 deployment tool it bridges the gap
7:26 between models and agents while
7:28 simplifying the complexity and
7:31 infrastruure Structure management model
7:34 is a complex algorithm trained a vast
7:37 train on vast amount of data. It learns
7:40 patterns and relationships in data
7:42 allowing it to generate new content,
7:45 translate languages, answer questions
7:47 and much more. What is infrastructure?
7:49 This layer provides core computing
7:52 resources needed for Genai. uh this
7:54 includes physical hardware like servers,
7:57 GPUs, TPUs and software needed to store
8:01 and run AI models and training data.
8:03 Come to the machine learning life cycle.
8:06 Uh the first part is data ingestion and
8:08 preparation. It is the process of
8:10 collecting, cleaning and transforming
8:13 raw data into usable uh format for
8:16 analysis or model training. And then for
8:18 model training uh what does that mean?
8:20 It's a process of creating your machine
8:23 learning model using that data. And
8:24 model deployment, it's a process of
8:27 making the train model available for
8:30 use. Uh model management, it's the
8:32 process of managing and maintaining your
8:34 models over time. Let's talk about some
8:37 of the Google offerings. Uh what is
8:40 Gemini? Gemini supports multimodal
8:42 understanding, advanced conversational
8:45 AI, content creation, and question
8:48 answering. What's GMA? GMA offers
8:50 developers user-friendly customizable
8:53 solution for local deployments and
8:56 specialized AI applications.
8:58 Imagine it's textto image diffusion
9:01 model that generates highquality images
9:04 from textual descriptions and vo it
9:07 generates video content based on text
9:10 description or still images.
9:13 Um now we will learn about gen Google
9:16 cloud generative AI offering and Google
9:19 uh genai tools are integrated across
9:22 Google uh Google's ecosystem and Google
9:24 ensures you stay updated with the latest
9:27 AI advancements. Google provides an
9:30 ecosystem that puts security and ethics
9:32 at the front line. Google provides an
9:34 enterprisegrade foundation that you can
9:37 build build on and Google open approach
9:40 gives you flexibility and choice in your
9:42 AI solution. There was a question about
9:45 this um uh
9:49 open approach in the exam.
9:52 Um Google tooling for personal uh productivity.
9:54 productivity.
9:57 Let's talk about Gemini. It's a Google
10:00 uh Genai model that pro powers many uh
10:03 different solution. The Gemini app is
10:05 Google's generative AI chatbot that
10:08 provides assistance with task such as
10:10 writing, summarizing, translating, and
10:13 creating images. Uh with Gemini
10:15 Advanced, companies can access extra
10:18 features and enterprise grade
10:20 protection. Gemini for Google Workspace
10:22 integrates Genaii into familiar
10:25 workspace apps, allowing you to do
10:28 things like compose email and Gmail,
10:29 generate images and slides, and
10:32 summarize notes and meet. Gemini for
10:35 Google Cloud is your AI assistant for
10:37 Google Cloud. It can help you write,
10:41 debug code, manage and optimize cloud
10:43 applications, analyze data in BigQuery,
10:47 and strengthen your uh security posture.
10:50 Beyond Gemini tools, uh we have notebook
10:52 LM. It allows you to upload your files
10:55 and then act as a research assistant
10:57 summarizing key points, answering
11:00 questions and generate ideas all while
11:02 staying grounded with your source
11:04 material. So there will be a question
11:06 about notebook LM. So it allows you to
11:09 upload your own content and then it acts
11:12 as a research assistant
11:15 and um it it stays grounded on the
11:18 material that you have uploaded.
11:20 Uh what is Vortex AI? It's a Google
11:24 cloud uh offering that um uh has unified
11:26 machine learning platform. It empowers
11:29 you to build, train and deploy machine
11:31 learning application. Uh Google AI,
11:35 Vert.Ex AI gives you access to genai
11:37 models such as Gemini and lets you tune
11:39 them to meet your needs and then deploy
11:42 them. Vertex AI search, search and
11:45 recommendation solution for uh business
11:49 uh use it with Gemini API with Google AI
11:52 studio or Vert.ex AI Studio. Uh Google
11:54 AI Studio is available free of charge
11:58 and is meant for quick AI prototyping.
12:01 Vert.ex AI studio is bu is for building
12:03 and deploying production ready AI
12:06 applications at scale. So there was a
12:09 question uh about this where like you
12:12 know how these two are different. Vertex
12:14 AI is uh studio is kind of for
12:17 enterprise level and this uh Google AI
12:20 studio is available for uh free of
12:23 charge and it's meant for you know quick
12:26 uh AI prototyping.
12:30 um customer uh engagement suite uh tools
12:33 to support your company in engaging with
12:35 customers effectively. They can be built
12:38 on top of Google's contact center as a
12:41 service CCA.
12:44 There was a question on this as well. An
12:46 enterprisegrade contact center solution
12:49 that is native to cloud. Uh
12:51 conversational agents act as effective
12:54 chat bots uh to your customers. Agent
12:57 assist support your live human contact
13:00 center agents and conversational insight
13:03 gain insight into all your communication
13:06 with the customer. So agent assist I
13:08 think this was one of the topic that
13:11 there were a couple of questions on it
13:14 was to support uh the live human contact
13:18 agent. Um, Gemini enterprise uh
13:20 integrate customized search and
13:23 conversational agent that can access and
13:25 understand data from various internal
13:29 sources into your organizations or
13:32 internal websites or dashboards. Tooling
13:35 uh there are like couple of things very
13:38 important again from the exam point of
13:40 view. What are extensions? They are used
13:43 to connect with external services via
13:46 APIs. What are functions? they uh define
13:49 a specific action or task. Data stores
13:52 provide access to the information and
13:54 plugins are like you know you add new
13:58 skills and integrations.
14:01 Uh Google cloud generative AI is
14:03 offering is continued here. Uh what are
14:07 agents? And um a geni agent is an
14:09 application that tries to achieve a goal
14:12 by observing the world and acting upon
14:15 it using the tools that are at its uh
14:19 disposal. Uh the components of an agent
14:21 is like a reasoning loop, tool and
14:24 model. A reasoning loop is an iterative
14:26 process where an agent observes,
14:29 interprets, reasons and acts often using
14:32 prompt engineering. Uh the tools are the
14:34 functionalities that allow the agent to
14:36 interact with its environment such as
14:38 accessing and processing or interacting
14:41 with the hardware and model is the brain
14:44 of the AI system which consists of
14:47 various algorithms that uh learn
14:50 patterns from data and can make
14:53 predictions or gen uh new uh generate
14:56 new uh content. Uh types of agent
14:59 deterministic, generative and hybrid.
15:01 Deterministic or also called as
15:03 traditional agent agents uh that are
15:06 built with predefined paths and actions.
15:08 Uh generative agents are that are
15:10 defined with natural language uh using
15:14 LLMs to give real conversational field
15:16 uh feel to your chatbot and hybrid
15:19 agents. They are these like combine both
15:20 deterministic and generative
15:23 capabilities and uh this is a
15:25 combination. It makes them more
15:28 powerful. agents can um serve several
15:29 different functions. Some of the
15:32 examples are customer service agent,
15:35 employee productivity, uh creative
15:37 agents, code agent, data agent, security
15:40 agent and I think there was one um agent
15:44 for uh uh Google maps uh that also came
15:46 um in the exam
15:50 uh platform the foundation for building
15:54 and scaling AI initiatives uh as Google
15:56 cloud's unified machine learning
15:59 platform vertex AI is designed to
16:01 streamline the entire machine learning
16:04 workflow. It provides the infrastructure
16:06 tools and pre-trained models you need to
16:09 build, deploy and uh manage your machine
16:12 learning and generative AI solution.
16:14 With Vertex AI machine learning op
16:17 tools, uh AI teams can better
16:19 collaborate to monitor and improve their
16:22 models model. Now, Vert.xi XAI gives you
16:25 options for how to handle AI models for
16:28 your project uh through model garden and
16:31 model builder. Model garden you can pick
16:33 from existing Google third-party or open
16:36 source uh models and model builder is uh
16:39 where you train and use your own models
16:42 and go fully custom to create and train
16:44 models at a scale using an machine
16:47 learning framework or use AutoML to
16:49 create and train models with minimal uh
16:52 technical knowledge and effort
16:54 infrastructure. It provides the core
16:57 computing resources uh needed for geni.
16:59 This includes the physical hardware such
17:03 as uh servers, GPUs, TPUs along with the
17:06 essential software needed to train,
17:10 store and run AI model, AI on the edge,
17:13 you can run uh AI solutions on
17:15 infrastructure devices or servers closer
17:18 to where the action is happening. And
17:21 Google uh provides tools such as uh run
17:24 uh light runtime to help developers
17:28 deploy AI model on edge devices and
17:31 Gemini Nano is uh Google's most
17:33 efficient and compact AI model is
17:36 specifically designed to run on devices.
17:39 Um now let's come to the Google cloud
17:43 generative AI offerings APIs. Uh
17:46 speechtoext API. The API convert speech
17:49 into text. It also transcribes audio and
17:53 video content. Textto speech API. It
17:55 converts text to natural sounding
17:59 speech. The API also creates voice um
18:01 user interfaces and personalized
18:03 communication. Translation API. The
18:05 translation API translates text,
18:08 documents, websites, audios and video
18:10 files. Document translation API. It
18:13 translates format formatted documents
18:16 while keeping the original layout.
18:20 uh document AI API. The document AI API
18:24 extracts data from uh the document and
18:26 automates data capture and document
18:29 processing. API can be used to summarize
18:32 document cloud vision API. This API
18:35 analyzes the image content tagging
18:37 images based on detected objects and
18:40 text. It can also identify faces and
18:43 landmarks. The API also supports uh
18:46 cases like content moderation and visual
18:50 search. Uh cloud video intelligence API
18:52 allows developers to analyze the video
18:55 content and extract meaningful info uh
18:58 content recommendation, video search and
19:01 media analysis. And natural language API
19:03 helps uh derive insights from
19:06 unstructured text, understand the
19:08 sentiment of the text, classify content
19:12 and extract important entities building
19:16 applications uh from your agents. You
19:19 can access Gemini API tool uh via tools
19:21 like Google cloud developer tools like
19:24 cloud run functions and cloud run and
19:26 low code and no code tooling like app
19:29 scripts and app sheet.
19:32 Now techniques to improve geni model
19:34 output. So they are prompting
19:37 techniques. Um zeros short asking the
19:40 model to complete a task uh with no
19:42 prior example. So in zero short you do
19:45 not give any example. Um one shot you
19:48 provide the model with one example to
19:50 learn from. Few short you provide the
19:54 model uh with a few or multiple examples
19:56 to learn from. and role. You assign a
19:59 persona to the model to influence its
20:02 style, tone uh and focus. Say for
20:05 example uh you ask uh in your prompt act
20:09 as a market researcher to do some to uh
20:11 something. So that is a role of market
20:14 researcher and prompt chaining engaging
20:17 in back and forth uh conversation with
20:19 the AI. So these are some prompting
20:23 techniques to improve the model output.
20:25 Uh let's uh talk about the reasoning
20:28 loop. Uh prompt engineering technique.
20:32 Uh react, reason and act. Allow the LLM
20:34 to reason and take action on a user
20:38 query. Uh chain of thought uh guide an
20:40 LLM through a problem solving process by
20:43 providing examples with intermediate uh
20:46 reasoning steps and meta prompting using
20:49 prompting to guide uh the AI model to
20:51 generate, modify and interpret other props.
20:52 props.
20:55 model guidance and refinement uh
20:57 grounding. There will definitely be a
21:00 question or a few questions on grounding
21:02 and retrieval augmented generation. This
21:06 is a very important topic. Um so uh
21:08 basically if you do not want uh your
21:11 model to hallucinate or it should only
21:16 use um the uh uh the content or the
21:19 knowledge that you have uploaded uh when
21:22 it is responding uh to your queries. it
21:24 is very important to ground it on that
21:27 data. So grounding is connecting the
21:30 AI's output to verifiable sources of
21:33 information and retrieval augmented
21:37 generation. Um you can say it um it it
21:39 comprises of four parts. First is
21:42 retrieval. The LLM retrieves the
21:44 relevant information from external
21:47 sources using tooling. Augmented
21:50 augmentation the retrieve information is
21:54 incorporated into the prompt uh to the
21:56 LLM generation the LLM processes the
21:59 prompt and then generates a response and
22:01 iteration it's optional the LLM can
22:04 iterate on a retrieval process as
22:07 necessary. So a person is uh sending a
22:11 prompt to the model the model queries um
22:14 the vector data and then it actually
22:17 responds um back. So when you do
22:20 retrieval augmented generation, it's uh
22:23 one of the way to ground and make your
22:31 Now let's come to streamlining uh
22:33 prompting workflows. Uh, reusing
22:35 prompts, saving prompts as templates for
22:37 repeated use, leveraging prompt
22:39 chaining, continuing conversation within
22:42 the same chatboard to maintain context,
22:45 using saved info in Gemini, storing
22:47 specific information for model to use
22:50 consistently. Gems. Gems are
22:53 personalized AI assistant within Gemini.
22:55 They use personalized responses tailored
22:57 to specific instruction. They also
23:00 streamline workflows such as uh
23:02 templates, prompts, and guided
23:06 interaction. Uh sampling uh parameters
23:09 uh setting that influence the AI model
23:12 behavior allowing for customized uh
23:15 results. Uh token count. This represents
23:18 meaningful chunks of text like words and
23:21 punctuation. Temperature. This parameter
23:24 controls the creativity or randomness of
23:27 the model's word uh choices during text
23:31 generation. Uh top p nucleus sampling
23:34 the cumulative probability of the most
23:36 likely tokens considered during text
23:38 generation. This is another way of
23:40 controlling randomness of the model
23:43 output. Uh safety settings. These
23:45 settings allow you to filter out
23:47 potentially harmful or inappropriate
23:50 content from the model's output output
23:52 length. This determines the maximum
23:56 length of uh generated output.
23:58 Now let's come to the techniques to
24:02 improve genai um AI model. Uh continuing
24:05 that the foundation uh model limitation
24:07 data dependency the foundation model
24:10 performance relies heavily on data.
24:13 biased or incomplete data will affect
24:15 their output. Knowledge cutoff AI models
24:17 are trained up to specific knowledge
24:20 cutff date meaning the uh there might be
24:22 uh lack of information about events
24:25 after that point. Bias LLMs learn from
24:28 large data sets which may contain biases
24:33 even subtle biases can have a magnified
24:36 impact on the model output. Fairness.
24:39 Assessing the fairness of geni ai models
24:41 is a key aspect of responsible
24:44 development. Hallucination when AI model
24:48 produces outputs that aren't accurate or
24:50 based on real information. This is
24:52 called hallucination. We talked about
24:55 it. Using grounding, we can prevent uh
24:58 some of the hallucination. Edge cases.
25:02 rare and atypical scenarios can expose
25:04 um models weaknesses leading to
25:08 unexpected um results.
25:13 Um so now we will talk about um human in
25:16 the loop. Uh the this is a process where
25:18 human input and feedback are directly
25:21 integrated into machine learning
25:24 workflows. Uh content moderation human
25:27 in the loop ensures user generated
25:30 content is moderately moderated
25:32 contextually catching harmful material
25:36 algorithm um might overlook. Sensitive
25:39 application human in the loop provides
25:42 critical oversight in fields like
25:44 healthcare and finance ensuring accuracy
25:47 and reducing risk from automated
25:50 systems. High-risk decision making uh
25:52 for high stakes decisions human in the
25:55 loop can help uh safeguard accuracy and
25:58 accountability through human review and
26:01 machine learning model output. Uh
26:03 pre-generation review human uh experts
26:05 review and validate machine learning
26:08 output before deployment catching errors
26:10 and biases before user impact.
26:13 Postgeneration review, continuous review
26:16 and feedback after deployment might help
26:19 machine learning model improve and adapt
26:22 uh to changing context and user needs.
26:26 Um additional terms context window the
26:29 amount of data um amount of text the
26:32 model can consider fine-tuning. It's a
26:34 technique used to enhance the pre-train
26:37 or foundation models performance for a
26:40 specific task or domains. Let's come to
26:43 managing your model. Um, Google Cloud
26:45 offers tools for managing the entire
26:47 life cycle of machine learning models.
26:50 This includes versioning. Uh, versioning
26:53 is keeping track of various versions of
26:55 the model within the Vert.Ex AI model
26:58 registry. Um, performance tracking, it's
27:01 uh to review the model metrics to check
27:03 the model's performance. Drift
27:06 monitoring, it's to watch uh for changes
27:08 in model accuracy over time with Vert.ex
27:11 Vertex AI model monitoring data
27:13 management it you it's used for um
27:16 Vert.x Vertex AI feature store to manage
27:20 and data features u the model uses the
27:24 storage um use vortex AI model garden to
27:28 store and organize models in one place
27:31 automate use vortex AI pipelines to
27:34 automate machine learning task now let's
27:36 come to business strategies for
27:39 successful genai solution before
27:42 starting your genai pro uh project
27:44 consider the
27:47 uh scale, how many users will be there?
27:49 Customization, how specialized is the
27:52 Genaii solution, user interaction, how
27:55 many users will engage, privacy, how
27:57 sensitive is the data, latency, what is
27:59 the response time you can have?
28:02 Connectivity, what is your connectivity?
28:05 Resources, people uh do you have access
28:08 uh to people who have AI expertise or
28:11 money? Uh what is your budget and time?
28:15 What is your project uh timeline? Genai
28:18 strategy uh there were two strategies
28:21 like top down and bottom up. So in
28:25 Google we uh Google recommends um a mix
28:28 of both uh it has to come from top down
28:32 and also bottom up. uh so it's uh a
28:34 combined approach a top- down and bottom
28:38 up uh approach where the leadership sets
28:41 a vision and a strategy and employees
28:43 identify practical applications by
28:46 providing feedback. Uh so what are the
28:50 advantages of this uh combined uh a geni
28:54 strategy? Uh it's u it has uh the focus
28:56 you pro prioritize focused geni
28:59 implementation with clear business value
29:01 exploration encourage experimentation
29:05 and collaboration to discover valuable
29:08 genai application. Responsible AI
29:10 establish ethical guidelines and ensure
29:13 secure and responsible AI development
29:15 resourcing. Invest in data strategy.
29:18 Leverage existing uh resources and
29:22 develop AI talent impact. Um measure GI
29:24 impact on business goals and demonstrate
29:28 tangible benefit. Uh continuously refine
29:31 um geni solution based on feedback uh
29:35 and data. Secure AI protect your AI
29:39 applications from harm. Um and using the
29:43 uh secure AI framework. It helps uh
29:46 organization manage AI ML model risks
29:50 and ensure security. Google cloud secure
29:53 by design infrastructure helps uh
29:57 support security across AIML life cycle.
29:59 Various uh tools help protect data
30:02 models and application. Identity and
30:05 access management IM for controlling
30:08 resources access. Security command
30:11 center for security posture visibility
30:14 workload monitoring tool to help build
30:17 and maintain secure AI system. Factors
30:20 when choosing a a model for your use
30:22 case modality. Choose a generative AI
30:25 model whose input and output data types
30:28 modality align with your application
30:30 specific needs whether it's text, image,
30:34 audio or video. context v uh window you
30:36 may need to balance an AI model ability
30:38 to generate coherent and relevant
30:41 responses against uh the increased
30:44 computational cost performance a model
30:46 accuracy speed and efficiency are
30:49 critical factors consider the trade-offs
30:51 uh between performance and cost
30:54 availability and uh reliability choose a
30:56 model that is consistently available and
30:59 performs reliably under load consider
31:03 factors like uptime guarantees redency
31:06 and disaster recovery mechanism. Uh what
31:10 is responsible AI? Ensuring your AI dev
31:12 applications don't cause harm and are
31:14 used uh in an ethical manner.
31:17 Responsible AI needs to be considered
31:21 throughout the entire AI life cycle from
31:23 uh data preparation and model training
31:27 and deployment to ongoing uh monitoring.
31:29 to plan for your Genai strategy.
31:32 Establish a clear vision, prioritize use
31:35 cases, invest in capabilities, manage
31:38 change, measure value, and champion
31:40 responsible AI.
31:44 Now, uh if you want to create your own
31:48 um genai uh study guide, you can go to
31:51 notebook LM, upload the study materials
31:54 there, and you will be able to create um
31:58 a study guide. you can uh create um you
32:01 know questions, quizzes uh and uh it is
32:04 going to help you and I hope uh this
32:09 overview of Genai leader was helpful and
32:14 you are able to learn and earn your
32:18 certification using this guide and uh
32:21 let us know in comment if uh you earn it
32:24 and we will be happy to celebrate and
32:25 congratulate you on your achievement.
32:28 ment. See you in the next video for
32:31 another U certification preparation. Bye.
