Generative AI for technology leaders | AWS Events
- Hello everyone, and welcome to AWS Innovate, Generative AI and Data Edition. My name is Tom Godden. I'm a director of enterprise strategy at AWS and former Chief Information Officer at Foundation Medicine.
Before we get started, I want to thank you for taking time to come and learn with us. From model customizations in Bedrock to vector databases in FMOps, today's event is packed with some amazing sessions designed to help you build with generative AI. And today I'm excited to share with you how you can innovate with generative AI. Wherever you are, whether you have a generative AI application in production today, or you're trying to move quickly to transform the way you engage with your customers or the way your employees get work done, you have the opportunity to change the way you do business. Generative AI has taken the world by storm because we've been able to, through customer facing applications, experience the most powerful and latest machine learning models. While a lot of attention's been given to how consumers are using generative AI, we think there's an even bigger opportunity in how businesses will use it to deliver amazing experiences for their customers and their employees.
We believe the true power of generative AI goes beyond a search engine or a chat bot. It will transform every aspect in how companies and organizations operate. In fact, Goldman Sachs forecasts that generative AI could lead to a $7 trillion increase in global GDP and lift productivity growth by 1.5 percentage points over a 10 year period. Gartner goes even further to say that generative AI is likely to be one of the most disruptive innovations yet encountered in the digital workspace, and that they expect it to impact 80% of jobs to some extent, with the information workers' job changing the most quickly and the most dramatically.
Customers around the world have deployed a wide range of generative AI applications and are now seeing the benefits of this technology and becoming more efficient in transforming the customer experience. Customers like Intuit. Over the past decade, Intuit has partnered with AWS, first to migrate their applications to the cloud and now improving their customers' experience with generative AI. Intuit's AI-driven strategy has led to the creation of GenOS, a proprietary generative AI operating system built on AWS. GenOS facilitates the rapid development and deployment of personalized financial solutions, and it does this in two ways, by providing the ability to harness underlying data and by providing access to a multitude of third party and large language models. These can be scaled out to customers with ease, all while balancing costs using Amazon SageMaker and Amazon Bedrock.
Intuit uses AWS to process an immense volume of data. Their systems are making more than 65 billion machine learning predictions a day. With GenOS, Intuit is building new applications like Intuit Assist for TurboTax, Credit Karma, QuickBooks, and MailChimp. In short, Intuit Assist is a generative AI-powered assistant that offers personalized insights to help users make smart financial decisions.
Another example is Adobe, where they're unleashing a new era of creativity with the development of Adobe Firefly. Since launching their first production model in March of 2023, users of Adobe Firefly, the company's family of generative AI models, have generated over five billion images, enabling users around the world access to powerful generative AI tools. Adobe moved quickly with generative AI, prioritizing speed and harnessing the power of their data. The professional quality content trained on Adobe stock assets and openly licensed public domain content has been an immense success with their Photoshop generative fill, where they saw 10X adoption compared to a typical new feature. Success stories like Adobe and Intuit are just the beginning.
Artificial intelligence is changing at a pace like we have never seen before, with new discoveries and innovations happening each and every day. At AWS, we've made it our goal, our obsession to lead the way for customers. Things like cutting edge innovations with our custom silicon to bring you the best infrastructure for your AI workloads, purpose-built managed services that provide a new wave of capabilities, and access to high performing models to make it even easier for you to securely build and scale generative AI applications. And of course, continued investment in training and resources, highlighted by our commitment to train 29 million people with free cloud computing skills training by 2025 through over 100 AI and machine learning courses. Pushing the edge of this new era requires a constant willingness to learn and take action. And in working with our customers, we've identified four foundations needed for you to help innovate with generative AI.
The first being, you gotta choose the right use case and you need to move quickly. Generative AI can accelerate productivity and transform your business operations in a variety of ways. When thinking about use cases, there are so many to choose from, but I like to think of these as front office and back office use cases. Front office use cases are those that directly impact the experience of your customers. Here's where you're using generative AI to reinvent the way they interact with your company and enhance their experience. The back office use cases, the ones behind the scenes, where generative AI is boosting productivity and creativity of employees or optimizing and driving higher efficiencies and lower costs with your backend processes.
But what does this look like in production today? The possibilities for generative AI to revolutionize your business and industry are nearly endless, with hundreds of use cases and opportunities at your fingertips. AWS customers of all sizes and industries are innovating in so many unique ways. Customers are continually working to enhance their customer experience, like the PGA Tour, who began collaborating with AWS and Amazon Bedrock during the summer of 2023 to delve into the possibilities of what generative AI can offer. The ongoing work is leading the PGA Tour towards the development of exceptional experiences across a diverse range of platforms, unlocking additional value for the tour, the players, and their fans.
Customers are also boosting employee productivity and creativity. Ryanair crew scheduling is currently managed by a team of operation planners at their headquarters, who are tasked with ensuring that planes and employees depart and return on time, taking into account training, holidays, and weather disruptions. In partnership with AWS, they created an app that helps Ryanair cabin crew manage their work lives in one place, using an uncomplicated tool in the palm of their hand. And companies like Adidas China are using generative AI for process optimization.
Adidas China wanted to improve their inventory management with visual components like backgrounds or models, so they looked degenerative AI for a solution. Together with AWS, they utilized product data to create a virtual try-on solution, generating lifelike models and proper backgrounds for Adidas products based on Stable Diffusion and controllable generative AI tools. But with so much choice, so much opportunity, how do you choose? How do you know where to start? When it comes to prioritizing use cases, the most important thing is to find what you can implement quickly and get building. You need to weigh the risks and requirements of each opportunity to get early wins with the technology.
By implementing these solutions, you can kickstart a flywheel of innovation within your organization. Early success drives excitement and buy-in, which makes more complex and more unique and innovative use cases all that more achievable. And that brings us to the second generative AI foundation, which is the importance of using your data to customize your generative AI solutions. When you want to build generative AI applications that are unique to your business needs, your organization's data will be your differentiator. When you think about it, every company has access to the same foundation models, but companies that will be successful in building generative AI applications with real business value are those that will do so using a diverse set of robust data.
You see, data is the difference between generic generative AI applications and those that know your business and your customers deeply. But how can you put this data to work? What data foundations do you need in place? First and foremost is having a data strategy for your organization. 93% of CDOs acknowledged the importance of an end-to-end data strategy, and its role in making their generative AI initiatives a success.
But along with the data strategy is also the quality of the data. The quality of that data matters in generative AI, because higher quality data improves the accuracy and the reliability of the model response. In a recent survey of Chief Data Officers, almost half of CDOs viewed data quality as one of their top challenges to implementing a generative AI strategy.
And data quality and data strategy are key because, well, data is growing at an incredible rate, powered by consumer activity, business analytics, sensors, and so many other drivers. That data growth is driving a flywheel for generative AI. Foundation models are trained on massive data sets, and then companies are using smaller private enterprise data sets for additional customizations of foundation model responses and learning and creating new intermediate data sets. These customized models will in turn drive more generative AI applications, which through more interactions creates even more data and even more data for that flywheel. And that customization, using your data to develop generative AI applications for your customers and employees is vital. Let's talk about three popular approaches for building generative AI solutions with your own data, and understanding which one is right for you and your use case.
Things like speed, accuracy, cost, and complexity are all factors. And you'll need to consider the business value for each situation and weigh the benefits of each to determine what makes the most sense. These three techniques are commonly ranked by their level of complexity from the easiest to the most complex. Many customers use prompt engineering. It is a simple, cost-effective process that lets you refine your inputs for generative AI so that you can get on-target outputs and optimal results. You can customize the outputs of an existing model with retrieval-augmented generation, sometimes referred to as RAG, without the need for retraining that model.
With RAG, the external data used to come from your augmented prompts can come from multiple data sources, including document repositories, databases, and APIs. RAG helps the model to adjust its output with data retrieved as needed from these knowledge libraries. Next is fine tuning. When you fine tune an existing foundation model, you're using a smaller sample from your own domain-specific data, basically creating a new model with your prepared dataset.
And finally, continued pre-training. Continued pre-training means you pick up where the foundation model provider left off, training the model on data sets in your enterprise to extend both the generalized and specialized knowledge of that model. And if you haven't tried using Amazon Bedrock to customize foundation models, well, you should.
It has a lot of capabilities and is evolving at a super rapid rate. And its supports all three of these capabilities to customize your model responses with your own data. With use case selected and data prepared, that brings us to the third foundation, building with the most comprehensive set of capabilities for generative AI. From startups to enterprises, organizations of all sizes are getting started with generative AI. They want to take the momentum they're building with early experiments and turn it into real world productivity gains and innovations.
At AWS, we're ready to help you reinvent with generative AI because we think differently about what it takes to meet your needs. And we reinvent again and again and again to help you deliver. We think about generative AI as having three macro layers, and they're all equally important and we are investing in all of them.
The bottom layer is the infrastructure used to train foundation models and run these models in production. The middle layer provides access to all of the large language models and foundation models you need, and to the tools you need to build and scale generative AI applications. At the top layer, we have applications built leveraging foundation models, so you can quickly take advantage of generative AI without any specialized knowledge. Let's start with the bottom layer, the infrastructure. With foundation models, there are two main types of workloads, training and inference.
Training is to create and improve foundation models by learning patterns from large amounts of training data. And inference uses those models to generate an output such as text, images or video. These workloads consume massive amounts of compute power. To make generative AI use cases economical and feasible, you need to run your training and inference on incredibly performant, cost-effective infrastructure that is purpose built for machine learning and artificial intelligence.
GPUs are chips that can perform a high volume of mathematical calculations simultaneously, making them popular for workloads like machine learning simulations and 3D rendering. If you're building your own models, AWS is relentlessly focused on providing everything you need, the best chips, the most advanced virtualizations, powerful petabyte scale networking capabilities, hyperscale clustering, and the right tools to help you build. Along with the infrastructure, you need to have the tools to build with large language models and foundation models. That's why we're investing in the middle layer of the stack. We know many of you need it to be easier to access a powerful and diverse set of large language models and other foundation models, and then to quickly build applications with them, all while maintaining security and privacy. And that's why we built Amazon Bedrock.
Bedrock is the easiest way to build and scale generative AI applications with large language models and other foundation models. Customers in every industry are already using Bedrock to reinvent their user experiences, their products, and their processes, and to bring AI into the heart of their business. Why Bedrock? You enjoy the broadest choice of models, many of which are available first or only on Bedrock. You can add your own business context quickly, easily, privately, and with the broadest selection of customization options. And you get enterprise grade security and privacy because we designed it that way from day one.
Customer excitement has been overwhelming. AWS is investing in the middle layer of the generative AI stack to make it easier for organizations to access powerful and a diverse set of large language models and other foundation models, and to quickly customize those models, all while maintaining security and privacy. We see a huge opportunity to help with that by infusing generative AI into systems people use in their daily lives. We believe generative AI should help everyone at work seamlessly.
That means helpful, relevant assistance, whether or not they know the first thing about foundation models, RAG, or any of the rest of it. And that brings us to the top of the stack, applications that you use that are powered by generative AI, leveraging foundational models. We believe generative AI has the potential over time to transform virtually every customer experience we know. It can pour through the nooks and crannies of a system and find data you never would've known had existed, and it can help you put it to optimal use.
It can generate insights that save hours of manual work, and it can take projects that were slogs and make them snaps. To bring generative AI to everyone at work, last year we released Amazon Q, a new type of generative AI-powered assistant designed to work for you at work. Q lets you get answers quickly with natural language interactions. You can easily chat, generate content, and take actions, all informed by an understanding of your systems, your data repositories, and your operations. And of course, we know how important rock-solid security and privacy are to your business.
Amazon Q can understand and respect your existing identities, roles, and permissions. If a user does not have permission to access something without Amazon Q, they can't access it using Q either. We have designed Amazon Q to meet enterprise customers' stringent requirements from day one. And we never use business customers' content from Amazon Q to train underlying models.
We've also announced Amazon CodeWhisperer. CodeWhisperer uses generative AI to allow developers to build faster, while freeing them up to focus on more creative aspects of coding. It uses a foundation model to radically improve developer productivity. It generates code suggestions in real time based on a developer's comments in a natural language. I have to say, there's few things as CIO that have made me stop and say, wow. But the productivity gains from CodeWhisperer do just that.
It's unbelievable. As we continue to reinvent for our customers, we've learned what is needed to bring generative AI to your customers and employees. You need the right capabilities to build performant, cost-effective infrastructure.
You need a secure, private, easy way to build and scale powerful new applications with foundation models. You need generative AI applications with capabilities that could be enriched with the context of your business. And it's all on AWS, built with our high bar for giving you broad and deep capabilities, choices, and enterprise readiness.
It's still so early in the game. And we're incredibly excited about what we can do together With the tools available to start building, it brings us to the fourth and final foundation for all of those using this technology. You need to take the steps to innovate responsibly with generative AI. Our commitment to develop generative AI in a responsible way is integral to our approach at AWS, and to do so, we focus on four key areas.
First, we want to help transform responsible AI from theory into practice, and help operationalize it across key elements of responsible AI. Second, responsible AI is an integral part of an entire end-to-end life cycle of a foundation model, including design and development, deployment, and ongoing use. It is not something that can be done in a silo, but instead it must be integrated across the lifecycle with a commitment to test, test, test, for accuracy, fairness, and other key responsible AI dimensions across all of your models. Third, you need to prioritize education around how generative AI works and what its limitations are. Finally, advance the science behind developing generative AI responsibly.
At AWS, we know that generative AI technology and how it will be used will continue to evolve, posing new challenges. Together with academic, industry, and government partners, we are committed to the continual development of generative AI in a responsible way. Altogether, we want to build generative AI that is safe, trustworthy, and a force for good. AWS deeply believes in responsible AI, and you'll see that belief in our products. I'll touch on just a few examples here.
First, transparency. Transparency means explicitly sharing the details and context to responsibly use your service. AI service cards are a form of responsible AI documentation that provide customers with a single place to find information on the intended use cases and limitations, responsible AI design choices, and the deployment and performance optimization best practices for our AI services.
We recently announced six new AI service cards, including Amazon Titan and AWS HealthScribe. And as part of responsible AI, AWS is providing intellectual property indemnity coverage for outputs of Amazon Titan models and Amazon CodeWhisperer. So if you use one of the generative AI applications or models from AWS that I just talked about and someone sues you for IP infringement, AWS will defend that lawsuit, which includes covering any judgment against you or settlement costs. It's part of responsible AI, and it's part of how we work. We are also investing in automating the adoption of responsible AI practices. We've announced a preview of Amazon Bedrock Guardrails, which is a new capability in Bedrock.
It makes it easy to implement application-specific safeguards based upon your responsible AI practices. Guardrails lets you specify topics to avoid and automatically filter out queries and responses in restricted categories. For example, an online banking application can be set up to avoid providing investment advice and limit inappropriate content such as hate speech, profanity, or violence. Another way to automate responsible AI is to build it into your foundation model selection.
With model customization in Amazon Bedrock, you can evaluate and select your foundation model based upon key responsible AI dimensions like accuracy, robustness, and toxicity. Customers can also use the same model evaluation capability in Amazon SageMaker Clarify. And that automated control also applies to our generated content from our Amazon Titan Image foundation model, which will contain an invisible watermark. A huge factor to enabling responsible AI is you.
Your goal is to be your own toughest auditor, so you're prepared for whatever comes your way for compliance in the future. AWS offers services like AWS CloudTrail, Amazon CloudWatch, Amazon DataZone, and Amazon OpenSearch Service. They're all designed to help you establish the right governance, the right auditing structure, and the right processes for your organization. Lastly, and maybe most importantly, when you use your data to customize generative AI for your business, your generative AI tools and services should be architected so your most valuable IP, your data, remains protected and private when you use customized foundation models to meet the demands of your organization. For example, Amazon Bedrock makes a separate copy of the base foundation model that is accessible only to the customer, and trains this private copy of the model with your data. This means your data is never exposed or used to train the original base foundation models.
Your data stays contained in your own isolated virtual private cloud environment, and it will be encrypted. And it means that you will have the same AWS access controls that you have with any other AWS service, with over 300 security services and features built to support you. We have covered a lot of ground here today on how you can find success on your generative AI journey. There is still so much more to learn and we are at just the start of this journey.
I think it's important to note that it's not just about how these models learn about our businesses through our data. It is how we as individuals and organizations learn as well. It's about learning generative AI techniques, learning to navigate the world of data transparency and responsible AI, and incorporating those learnings about this new technology to help our businesses, our users, and our communities. At Amazon, we have a leadership principle called Learn and Be Curious. It says that we are never done learning.
We're always seeking to improve ourselves. And we are curious about new possibilities and act to explore them. This event was made for those of you who are curious, those of you who want to learn, and those of you who are ready to unlock the power of data and generative AI. We're excited to see what you build and AWS is with you every step of the way.
Thank you.
2024-07-07 04:09
