- 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.
