Choosing the right generative AI use case

Good afternoon, everyone. Thank you very much for attending this session. I know it's late - 5:30 in the afternoon, so we really appreciate you coming to this session. Thank you.

My name is Alvarez Luga. I'm the Head of AI Solutions Marketing at AWS and I have the pleasure to share the stage with me today with Kali Heath and Sama Bali, Senior Product Marketing Managers at AWS who will join me on stage later.

This is our agenda today:

• I will first give an introduction of what are the top generative AI use cases we're seeing among our customers.

• Then we'll do a deep dive in three of those: chatbots, employee assistance, and data augmentation.

• And then I'll come back on stage and give you a kind of recommendations and best practices based on what we're seeing with our customers and in the Generative AI Innovation Center.

So let's start by talking about generative AI. I'm sure that you have all experienced the transformative power of this technology through consumer facing applications. But we are convinced that even if from a consumer perspective, it's an incredible technology, when we move these in the business space in the context of your organization, where this technology will really thrive and bring value to your customers, bring value to your employees and optimize business processes in the back end.

In fact, Goldman Sachs forecasts that in the coming 10 years, gen AI will bring $7 trillion in GDP increase in the next 10 years. So huge transformation, transformative technology.

Let's have a look at what makes this technology so special. Traditional ML is based on using large amounts of labeled data and training models. And then you have a model that is specialized for a specific use case - forecasting, term prediction, predictive maintenance - large amounts of data, label training a model that is capable of doing one use case.

Large language models and foundation models are trained with large and huge amounts of data that is not labeled and are capable of doing many different tasks. That's one of the main differences between the traditional ML and gen AI and foundation models.

You can do things like generation, creating new text, creating new image, creating new music. You can do question and answering. You can do summarization, take a large document and summarize it in just one paragraph. You can do translation both from a language perspective, also from an audience perspective - I want these social posts for a more younger audience, for a more traditional audience, more from one concept of one industry, another industry.

Correction of content and also classification - those are the six main use cases that foundation models can do and they can do it in multimodal foundation models with different inputs and outputs. You can have text as an input and text as an output, text as an input, an image, as an output, image as an input and text as an output. And all the different combinations with text, image, video and audio and many more. It could also be molecules kind of the output that you're creating.

And then finally, when you combine those six capabilities with those inputs and outputs and with all the different industry needs, you end up with hundreds of generative AI use cases. There's hundreds of generative AI use cases across the different industries.

All those generative AI use cases could be grouped in those three categories and I usually present them that way - who's gonna benefit from that technology? Is it gonna be your customers because you're gonna use generative AI to enhance the way that you engage with your customers? Is it gonna be your employees who will benefit from generative AI because you will be able to boost their employee productivity as well as your creativity? Or is it gonna be more your back end process, your IT process, the optimization and using generative AI as a way to optimize your business process?

Those are some examples:

Chatbots, we'll do a deep dive with chatbots, virtual assistance when you add voice and transcription could also be a voice of the customer, conversational analytics, being able to transcribe the conversations in the contact center and then summarize and extract the key insights from those conversations - what is happening, what the customers are talking about, what are the requirements from a product perspective, what is the feedback that you receiving? It could also be what we call hyper personalization - so not just the personalization, a kind of segmenting, who's the segment for a specific product and what is the next recommended product - but more about how to have a personalized communication with our audience using generative AI.

In the employee productivity and creativity, the main one is the concept of employee systems using generative AI systems to help your employees in the context of your organization be more productive. But you can also use it for developers for code generation. You can also use it for business analysts to be able to create new graphs, new visuals, new stories with data and many more.

And then on back end optimizing business process, there's traditional use cases like IDP - intelligent document processing. All the organizations manage large amounts of documents and document processing is the capability of extracting data from those documents and then processing that information in your back end. But now with gene AI, you can also do summarization and classification of that information. You can also use it for training cybersecurity models, fraud detection models. You can do it for pure process optimization and something that Sama will talk about later, which is using gen AI to create synthetic data for data augmentation.

Those are just a few as I was mentioning, there's many more. That's when we look from a horizontal point of view. But if you look from an industry, then new use cases come up and these are just some examples of them.

But in healthcare, you have AI-enabled digital scribes being able to listen to the conversation between the patient and the doctor and then do a summary, do create the document that will be uploaded to your EHR. Be able to use gene AI as a way to improve the quality of medical images for detecting cancers or detecting whatever health issue.

Drug discovery - that's more about generating new molecules for drug discovery. It could be applied for in the industrial space for maintenance assistance. So helping your technicians based on all the documentation that we have about the different assets in your manufacturing plan, training the agent and being able to teach and to help the technician in order to fix that specific equipment.

It could be used for product design, also used for product gen AI to design products. It could be used in the financial industry for fraud detection or for intelligent advisories - that you're helping your customers using generative AI in those conversations and suggesting what are the recommended investments or the recommended financial advice.

Retail, we can use this for example for summarization of feedback that you're receiving from products. This is something that Amazon.com is doing - so you have all the reviews from all the different products and then you're summarizing all those reviews in a single paragraph or one page document that explains all every single feedback and identifies what are the key points from the feedback perspective.

Or you can also use it by taking an image and generating and suggesting what will be the description of that product in a way that it will engage - so improve the engagement that you have in the products on your ecommerce platform.

Media and entertainment - creating high quality documents at scale. It could be used for improving existing content, media, video - improving the quality of videos from a resolution point of view, from a framing point of view.

So it can be applied as you can see across every single industry. Let's have a look at those - those are kind of overview of the main ones. Let's have a look and a deep dive on some of them.

And let's start with chatbots. So I would like to introduce Kali Heath who will introduce all the chatbots and in deeper detail.

Thank you. Good evening. Thank you all for being here today. I'm very excited to share the stage with my colleagues today and talk to you all about conversational AI and generative AI and how companies of all sizes and backgrounds can leverage that technology to improve the way they serve their customers.

I'm gonna focus on self-service chat bots. But when we talk about conversational AI, we're also talking about IVR systems and virtual assistants. So on this end, within AWS, our customers are starting to really get excited to find new ways to augment their conversational AI interfaces with generative AI to make those conversational chat bots more of a natural sounding language, make it more conversational and make it in a way that's that people can interact with it in a natural sounding way.

We're also seeing recent trends, analyst reports back that up as well. In fact, the recent Gartner study shows that about 80% of conversational AI applications will have generative AI embedded in it by 2025. That's up from just 20% this year in 2023. And so it's safe to say we'll see, you know, an uptick and even a proliferation in the use of generative AI for things to help with things like mundane tasks. So think about information retrieval, automation of workflows and even summarization.

The benefits that customers see from leveraging chat bots in conjunction with having customer care teams is also pretty compelling, right? So if you think about a customer contact center, you'll start to see your call volumes reduce. You'll start to see a reduction in call abandonment rates. And for the customers that are engaging with your products and with your brand, they're able to interact in the different channels that are their preference, whether that's through voice through chat. So they, they're being provided with omni-channel access to your brand 24/7 that is provided at a much lower cost right from an operational perspective than having 24/7 support from just humans.

That also gives your customer care team a focus to just focus on issues that really need a human touch, right? So you're offloading the basic repetitive questions and Q&A, issues that can be handled by a chat bot. You're offloading that to a chat bot while, while you know, unburdening your customer care teams.

So when we think about deploying conversational interfaces, it just depends on what you're trying to achieve from a customer experience perspective that will determine what type of functionality you'll need out of these chat bots, right?

So on one hand, you have traditional chat bots, they're really, really good at particular things when we think about predictability, when we think about control, right? If you need predetermined responses, if you need not a lot of variability in what that chat bot is spitting out, maybe you're in a very highly regulated industry you want to go with this type of chat bot, right? These are basic FAQ chat bots and this really gives you predetermined responses and it'll have that control in place for structured responses.

But what we're starting to see, our customers demand is more dynamic, personalized responses coming from those chat bots, right? So that's when you get into this space around generative AI strengths. Large language models are really, really good at other things, right? So if you think about it, large language models are open and they're exposed to a lot more real world data, they can pull from domain specific data and they can even pull from trusted knowledge sources within your company.

That is really where you get the lift and the value because now you can get responses that are conversational, they're maintaining that context, they're maintaining that dialogue and they're responding in a more dynamic and more creative way.

So what we also see is an opportunity for generated AI to improve the bot building process. By a show of hands, I just wanna see how many people have leveraged Amazon Lex to build chat bots. That's one of our AI services and oh, that's good. That's good to see. I'm glad you guys are supportive of that service.

So what we found with our customers is that they're running into a cold start problem. Developers are spending hundreds of hours having to be bogged down by analyzing transcripts uploading historical conversations and trying to pinpoint the intent, the utterances and to think through and anticipate all the different combinations of questions and answers. This takes a tremendous amount of time and it's prone to human errors and it would actually be much easier if they could start off from a conversational flow template versus starting from scratch when building these bots.

That's where generative AI plays a huge role. It comes into play within the design and build process of the bot building whole experience for you just this week earlier, this week. During re:Invent you may have heard of a few of the generative AI announcements that are linked to Amazon Lex. Very exciting advancements being made to help developers accelerate the time it takes to build these chat bots.

So instead of taking, you know, thousands of lines of transcripts, uploading that and predefining what that intent is, you can simply engage with the chat bot in a very conversational plain, you know English or whatever your language is way to describe that chat bot and Amazon Lex will build that for you.

So for instance, if you have a customer use case where you wanna build a bot that allows your customers to take orders, a food delivery order, and you want that bot to also accommodate changes to the order, maybe cancellations to the order - you would just simply give the Amazon Lex that simple prompt in natural conversational way and it would go and build out the sample intents, utterances, prompts - in fact, it will kind of build out an entire conversational flow for you to visualize how that bot could come to life.

To me, that's a really huge, you know opportunity where builders can save time to, to get these bots to market quicker.

So now we'll just take a few minutes to look at a couple of different customers who are doing just that - they're out there leveraging these foundational models in addition to some of our conversational AI capabilities to better serve their customers.

Booking.com - huge startup company, big brand name started out of the Netherlands and they've quickly grown to become one of the world leading, you know, digital travel agencies around the world. They are on a mission to break down that friction that's included in booking travel.

Today, it's pretty cumbersome to book travel, right? There's just so many things you have to consider between destination and price and dining experiences and excursions. And it's almost as if we wish technology would just connect those dots and provide a more personalized experience for us.

Well, that's what booking.com is doing. They're leveraging Amazon Bedrock and they're customizing large language models with their proprietary data, by the way, that is a huge differentiator for them as a business, right, where they can take data that they have on their customers on that specific travel market and layer on different insights for that customer.

So by using Amazon Bedrock Booking.com is delivering a more personalized experience through conversational chats.

Next, I'll point to Quest - it's a mid-size company. It's not as large as booking.com, but also a digital first company who focuses on providing communications between residents and property management companies.

They've seen an opportunity in the market to do this in a more conversational way. So they're working with Amazon Bedrock, the Amazon TITAN model, in fact, to deploy state of the art virtual assistants that are very conversational in manner.

They understand that they need residents to or excuse me, the renters to be exposed to the most relevant data, up to date on the market, different neighborhood amenities, what's available price points, that type of thing.

These property management companies also want to provide that information for them, but they need to do it in a way that's not cost prohibitive. And so providing 24/7 support through the channels that they prefer to engage their customers, right? They're doing it through email, chat, voice, text. This is providing omni-channel access, which is what I mentioned earlier as a way to improve the communication and the engagement that they're seeing from their customers.

Lastly, PGA Tour - it's a nonprofit, a professional golfing association based in North America.

But they have golfers from 28 different countries around the world. They've been on a mission in partnership with AWS to help transform their data for a couple of years now and are looking to leverage new generative AI capabilities to elevate the fan experience even more.

They're experimenting with things like the retrieval, augmented generation prompt engineering and even vector databases to bring about new connections for their customers. They've actually leveraged the AWS Generative AI Innovation Center to spin up very quickly a proof of concept on a virtual agent that provides again conversational um capabilities, not only for their fans, right? So if their fans want quick access to insights on players and tournaments, they're also providing that information to the broadcasters so that they're able to stay on top of having that information at their fingertips.

These are just three examples of customers that are hitting the ground running, trying to augment their existing conversational AI applications with generative AI. And the whole goal behind that is to create more natural sounding conversations for their customers because that that is where they see the benefit of having this technology, this transformative technology at their fingertips to help improve the customer experience.

Lastly, what you're looking at is an eyesore I know, but it's, it's actually the Q&A chatbot solution. It can be found in our AWS solutions library and this is really how you get started with a lot of the things we've talked about today, right, augmenting your conversational AI interfaces um with bedrock capabilities with large language model capabilities. This is a CloudFormation solution that you can launch within the console and it allows you to deploy um self service omni channel, chatbot solutions, no code, no coding experience uh required. So this is really how you're able to put the right controls in place, have your data layered on to these large language models in a very secure manner and still deliver it through the channels you want, whether that's through the contact center, through voice through text applications.

So I think what I'd like to leave you with is that uh we encourage, you know, the customers, the developers out there to get started experimenting with these different applications and ways you can improve the customer experience. I would also encourage you to look into the recent Generative AI launches from Amazon Lex announced just these over the last couple of days. And with that, I'd like to turn it over to Miss Sam A Bali who will talk to you more about how generative AI can improve employee assistance as well as improve data processes. Thank you.

Good evening, everybody. I'm S Bali. As you can see, I'm tagging team with my colleagues from the AWS product marketing team for machine learning and generative AI. And I'm here to really talk about employee assistance. And I think all of you here are going to agree with me that in today's workplace, the most critical thing is productivity and generative AI offers immense potential, especially when it comes to boosting employee efficiency.

So in this section, I'm really going to be walking you through how generative AI can be the guiding light for the future of efficient work. So now most activities today at a workplace are routine, repetitive activities, right? You've got scheduling meetings, data entry, processing your documents, answering routine customer questions. In fact, a recent study by Mckenzie from this year calls out that these routine activities just on them, an employee spends about 60 to 70% of their time and they can now be automated by generative AI.

The same study also calls out that employers who are now really investing in generative AI for the employees are seeing benefits in four key areas. So starting first with customer operations, that's where what are really covered right now, where you have these generative AI powered chat bots, you have virtual agents which can really help you transform your customer experience. Second, you've got marketing and sales. So now you can use generative AI to create targeted emails, ads campaigns at scale and then continue to use your own data to really optimize these campaigns.

Third, you've got software engineering. Now you can use and rate of AI to boost developer efficiency with auto generated code as well as automating a lot of the work flows. And then lastly, you've got R&D and this is applicable across all industries where you generative AI, you can really accelerate discoveries by surfacing novel hypothesis from your data.

So now let's talk about how are these organizations bringing generative AI together for their employees? And it starts with an employee assistant. This isn't just a chat bot. It's powered by groundbreaking AI. It understands context. It can have natural conversations, think of it as an ally to really amplify the human potential and it takes care of your day to day tedious tasks.

Now, we've got four key areas where we really seeing the benefit of an employee assistant. First is content discovery. I would love to see a raise of hands for folks who spent more than 15 minutes looking for a simple piece of information. Every time you've joined a new company or a new team. I know I'm one of them. There you go. So with these employee assistance powered by generative AI, you can now connect all your keys, all your share points, your sales force, your confluence, any other application to really offer a unified search experience?

Next, you can really help your employees have better decision making processes. You can unlock insights from data within your organization and help them have a faster but a data driven decision making process. You can use generative AI to summarize large amounts of data found in your organization and then conduct competitive analysis. And then lastly, content creation. I'm a marketer. I've been a marketer all my life. And if I had a dime for every time, an engineer or a product manager asked me to create a social media post for them, I would be in a very different place right now. But hey, now you all have generative AI, right? You can empower all your employees from sales to marketing to HR to engineers. They can all now create content with generative AI across all platforms.

Let's get a bit deeper into using generative AI for each of these different functions starting with sales where you can use generative AI to create personalized messages, emails, talking points, all while specifying a specific customers, challenges, pain points, the requirements, what the competitor market looks like for marketing. We spoke about content creation, but being a marketer myself, I know the time expense and how difficult it is to create an identity for your organization.

Now we generate over here, you can create logos, color patterns again all within the defined set of parameters. Next, you've got product managers, you can really distill customer feedback and create new designs, optimize existing designs or even figure out prioritization for the next set of features you're going to deploy upper management can use a rate of AI to really look at insights across the organization to come up with faster. But again, a data driven decision making process, the same goes for organizing teams for better collaboration.

Legal can use generative AI to summarize to get insights from large legal documents, they can even create contracts agreements, letters all based on a specific template with a user input. And lastly, HR not only can you use generative AI to create new job descriptions, you can also use the same technology to analyze resume a submitted against it. Similarly, you can create on boarding plans which are posterized to employees, right? Depending on their job skill, depending on their geographic location, depending upon the time they need to be doing the training.

So moving forward to organizations who are making these employee assistance real for their employees starting first with adidas, this is a global footwear athletic app organization. Now they've created a generative AI solution for their community of adidas engineers. This is a search interface which lets them find information answers for everything starting from getting started to highly technical questions and it all comes together in the form of a conversational search interface.

Next i have is crea now this is a consulting firm. They've created a generative content accelerator using amazon bedrock. So now they're using LMS to analyze any kind of marketing brief and then they're creating a lot of these phrases which can be used for your campaign. They're localizing it for you. They're also creating prompts for images that can be used in the campaign.

Next, I have Lexis Nexus Legal Professional. So now they've created Lexus Plus AI system which really features intelligent search summarization and legal drafting capabilities to make lawyers be more efficient and effective.

So one thing you'll see across all these stories is a key to him. The idea is to use generative AI to help you take care of your day to day jobs and then make your employees be more engaged, be more productive, be more empowered.

No, when it comes to deploying these employee assistance for your organization, you've got a couple of options. One, you can be building your own solution using Amazon Bedrock or you can be using out of the box solutions like Amazon Q. Now this is hard of the press. This is what Adam announced today in the keynote as well. So Amazon Q is our generative AI powered assistant. It's designed for work and it's tailored for your business.

So now you can have conversations, you can generate content, you can solve problems and you can take actions using your data and expertise found in your company's repositories, code and enterprise systems. When you are having a conversation with Amazon Q, it's giving you ex relevant imaged advice and response to help you really streamline work flows, speed up decision making, but most importantly, really help you spark innovation and creativity at work.

Now we've built Amazon Q with security and privacy in mind, which means it understands and it respects existing identities as well as your roles and permissions. So let's say you have a piece of information, which one particular employee could be me who does not have access to. So let's say I don't have access to payroll information. Even when I'm using Q, I will still not have access to the same piece of information.

We've also designed Amazon Q to meet enterprise stringent requirements. So none of your data or the input and output prompts from Amazon Q are we're not taking any of that to train the underlying models for Q for anyone. But you now we've also decided to design Amazon Q in a way that it's available everywhere you're working.

So it's available as part of your business. But Amazon Q is also trained on AWS information over the last 17 years. So we've made this your AWS expert. So as you are building, as you're deploying operational, your workloads, your applications. AWS Q is part of the Amazon Q is part of the AWS management console, your IDE through code whisperer, as well as your Microsoft chat rooms, your slack rooms and your documentation. So this is a good way to have a conversation on with Q on AWS and find new solutions, new architectures best practices. Think of this as your expert on AWS.

Amazon Q is also available as part of Amazon QuickSight. So that's a business intelligence service. So let's say you're looking at a particular dashboard and QuickSight and you're wondering why did my sales go up last week? You ask that question to Amazon Q. And in response, you get a mini dashboard with the factors that illustrates the changes that really triggered the sales up from last week.

Business analysts can also have a conversation with Amazon Q and ask you to create a new dashboard based on certain data they want to use. Now, we have also made Amazon Q available with Amazon Connect. So that's a contact center service. So your contact center agents can now have a conversation with Amazon Q while they're chatting with the customer to get responses quickly.

Similarly, we've also made Amazon Q part of AWS Supply Chain. So that's a supply chain insights service. So supply chain managers, inventory managers can now really dig into stock out or over stock risk and get to creating a mitigation plan much sooner.

So now i have a really nice surprise for everybody. You all are probably the first ones who get to see Amazon Q in action. So I'm going to keep quiet here and let you all see.

Alright, I will let the product team know that all of you were definitely impressed.

Ok. So I think up to now all Albert Curly and I we've been talking about these use cases where you can apply generative AI, I'm gonna shift gears a little bit and talk about how you can use generative AI to get more data to train your machine learning models. Because I think all of you here will agree that your machine learning models are only as good as the data they're trained on. And in a lot of real life scenarios, there is scarcity of this large labeled data set to really power your generative AI applications, especially when it comes to medical imaging, autonomous driving. And this scarcity leads to a machine learning model with limited performance and really less generalization capabilities. And that's where data augmentation comes into place in a very simple form

"Data augmentation is the process where you're creating more training data from your existing training data. And the idea is that now you're exposing your machine learning model to a larger data set, a more varied and diverse data set to increase and improve its performance.

Let's talk about where you really require the augmentation. So it really comes down to three simple options and factors.

First, you require the augmentation where data set is small. So let's say you're creating a pedestrian detection model for a self driving car. And you realize, ok, i don't have enough images. I'm gonna go start taking pictures of pedestrians crossing roads and walking around and you'll soon realize that collecting this data is not easy and you've only ended up with maybe 100 and 150 images. Whereas with synthetic data, you can create thousands to millions of images in a fraction of time.

Next, you require data augmentation where you have imbalanced data sets. So that means that not all classes in your data set are represented equally. Let's say now you're creating a fraud detection model for a banking system. Here, your legitimate transactions are going to be much, much higher than your fraudulent one, getting you an imbalanced data set.

And then third, you require data augmentation where your data set contains sensitive information. Now you're creating a model which is going to predict employee attrition rate, which means now you have sensitive details pertaining to your employees.

So that's these are the three scenarios where you require data augmentation. And we expect that most of the AI systems of the future are going to have a combination of real data as well as synthetic data. And because it's because of a few factors, right, first being in most scenarios, especially when it comes to health care, autonomous driving, it is next to impossible to gather large amount of data to train your machine learning model.

Second, even when you are collecting the data, there are concerns around data privacy, you are now liable to collect and store this data for a large period of time. And then lastly, we spoke about imbalance data sets where now with synthetic data, you can easily balance these q data sets and make sure that these unrepresented classes are also represented correctly. The combination of the authenticity you get with real data along with the flexibility as well as control you get with synthetic data. That's what gets you the best of both worlds.

I'm going to now move on to a story really describing how Amazon One was using data augmentation. So Amazon One is a fast convenient contactless experience, whether you're using the palm of your hand to do a variety of things. It could be to pay at a store, it could be to show a loyalty card. It could be to verify your age, it could be to enter third party venues as well. Now, Amazon One is currently being rolled out to hundreds of Whole Foods as well as thousands of third party venues through a 12 partners. In fact, if you stop by a Hudson store at one of the US airports, you might actually see one of these.

So the way Amazon One works is they have an infrared scanner and it scans the palm of your hand. So it's looking at the unique lines, the grooves, the ridges in your palm. At the same time, it's also looking at the pulsating network of nerves right behind the skin. So once it scans the palm of your hand, it denotes a unique numerical vector to it, which is then connected to your credit card, your Amazon account.

Now the success of Amazon One depends on accuracy. There is no room for error. So now the team behind Amazon One started to train the machine learning model. But when was the last time that you saw pictures of hands, right, not on Instagram or TikTok? If there was that, I definitely missed something there.

So they turned to data augmentation and then they created millions of synthetically generated pictures of just the palm of your hand with slight modifications, right? Different lines, some hands with wedding band, some with band aids. They even trained the machine learning model to detect artificially generated silicon hands to improve and boost its accuracy.

And I'm really excited to call out that Amazon One at this point has been used more than 3 million times with 99.9999% accuracy. That's four lines of accuracy rate. It's also 100 times more accurate than your iris.

Another story which I would like to talk about is Stick Mass. So this is an online education provider and they provide personalized learning experiences for students of all skill levels. Now, their mission is to really understand every learner at an individual basis and then adapt content learning resources as for that.

So Tick Must turned to our AWS Generative AI Innovation Center and they wanted a chat bot. Now this chat bot should understand the skill and learning level of each student. It should also understand the emotional state of that learner and at the same time should be able to answer a myriad of questions. But the issue was that Tick Mas themselves did not have access to any questions or a way to really evaluate this chatbot. Either.

So AWS Generative AI Innovation Center created thousands of synthetically generated questions all in accordance with the learning content with parameters such as emotional level, the reading level of students, the skills of students which could be controlled. These questions were then vetted by the SMEs at the Mass to really create a good data set which is now powering their chat.

Now, data augmentation can be used for a variety of use cases. It can be used for images to really help you balance and get more and more data for your data set. It can be used for videos for audio for text. You can also use data augmentation to inject anomalous behavior to create an anomaly detection model. But the biggest benefit for data augmentation comes from the fact that it helps you foster responsible use of AI one. It's helping you balance that data set. So it's helping you reduce bias next. Since you do not require that many amount of real data, you're reducing your privacy risks, you don't have to store it and manage that data for a large period of time. Whereas with synthetic data, which does not contain any private details, it still provides the signals required for the machine learning model.

Third, it's helping you improve accessibility because synthetic data really reduces the computational needs and the data costs making it accessible for a variety of use cases. And then lastly, you're also improving your model robustness. Now your data, now your machine learning model is exposed to a large amount of diverse very data, helping you improve performance. And at the same time also promoting best practices for using data for machine learning.

In fact, AWS is also on a mission to ensure that we are building generative AI in a responsible way. We are focusing on four key areas.

First, we are ensuring that we transform responsible AI from theory to practice for this. We are providing tools, guidance and resources to ensure that not only are you building generative AI in a responsible way, you're also using it responsibly.

Second, we're integrating generative AI end to end in the creation of a foundation model starting from design to development to deployment all the way to ongoing use. And we understand that that cannot be done in a silo. And for that, we're committing to testing, testing and more testing for fairness and accuracy.

Third, we're prioritizing education around how generative AI should be used. Just last year, we introduced AI Service Cards. These are your one stop shop transparent resource to find information about every AI service from AWS. It tells you about use cases, benefits limitations and also responsible AI designs.

And then lastly, we understand that as the science behind generative AI advances. There are going to be challenges which we will have to mitigate, but we will commit it. We are committed to really work with our partners in government academic as well as industry to ensure that we continue to develop generative AI in a responsible way.

So with that, I am going to take a pause here and invite Albert back on stage to give you some tips, tricks on how to get started with the generative AI project. Thank you.

Thank you very much. Good news. And the last speaker, we don't have more speakers and I just wanna close this session by giving some advice and some recommendations based on the experience from the Innovation Center.

I'm sure you have all seen on LinkedIn. That thing that says the answer is generative AI. What was the question? I think that we're a little bit in that situation. We all have CEOs that are saying, what is our generative AI strategy? And what is that we're doing?

One of the main advice is to focus on what is the problem fall in love with the problem, not the solution. We know that generative AI is exciting. We wanna try, we wanna find how we can use it. Let's first find what is the solution that we're trying to solve? What is the problem? Something that should be with a very specific goal that could be measurable that will have an impact, something that is unique to something that is trained with your own data.

If you end up deploying a generative area like everyone else, there's not going to be a differentiation. Yes, you've got to play with generative AI, but there's no differentiation. There's no your data. The only way to have something that is unique for you that will be a competitive advantage is by having foundation models that are fine tuned through the prom through rack, through fine tuning with your own data.

And then second, once you have identified, what is the use case, what is going to be the way that you will measure the value, then start a process in which it gonna be iteration, multiple iterations until you get it. right.

Second, you also want this to be the first of many projects within your organization. So try to find something that is impactful, something that will be visible within the organization or because it's visible or because you wanna make it visible by communicating and then something that is easily relatable so that the rest of the teams, the rest of it within the organization will say ok, that was a fantastic example, a very strong business use case, something that i understand and that can relate. And i can find out what will be that use case in my own department that we could apply this.

If you want to find additional use case, both AI and generative AI we launched last year, last year, the AI Use Case Explorer, very easy to use tool in which you just select what is your industry, what is the business function and then what is the business outcome? And then the tool will suggest some different use case. And for each use case, you have example of customers example of solutions and, and, and how to implement those particular use cases.

The second one is, and Kalli mentioned this before, we have the AI Solutions in the Solutions Library. You have a bunch of CloudFormation solutions for AI and for generative AI that you can download to the console and not start from scratch.

And then finally, we also have the Generative AI Innovation Center for those organizations that are interested in working with AWS in order to start their innovation projects. They've been working with many different customers. We also have our partner ecosystem and we have a lot of partners that are working also with Bedrock and with all our generative AI technologies that will be more than happy to help you.

The goal of this session was just to show you examples of many different use case, many different customer stories. The main output will be if you find something that you've seen today that say, well, maybe that's a use case that i could apply in my organization and then follow that kind of guidance on focusing the solution and the impact rather than just the technology.

Thank you very much for attending this session. Thank you very much, Kalli and S for attending. Thank you."