“AI chatbots are what everyone thinks about when we talk about this new branch of machine learning (ML) known as generative AI, but there’s actually so many more applications of this technology,” explains Steve. “We’ve been using ML for over 20 years at Amazon, from the first book recommendations on our website through to path optimisation for robotic picking in our fulfilment centres, supply chain forecasting, Alexa, and the computer vision technology in Amazon Go – our retail experience that lets consumers select items off a shelf and leave the store without having to physically check out.”
Generative AI is a subset of ML that can create new content and ideas, including conversations, stories, images, videos and music. It is powered by ultra-large ML models, also known as ‘foundation models’ (FMs), including large language models (LLMs) and multi-modal models (for example text, images, video, and audio). Examples of FMs include Llama 2 which was created for language-based tasks, and Stable Diffusion, which can generate images from text prompts. FMs have captured global attention in the last few months, and not without reason.
When working with FMs, you may need multiple different types during a project, and extensive ML training, raising the entry requirements for businesses to innovate with AI. However, Amazon Bedrock – a brand-new, fully managed service from AWS – allows developers to build and scale generative AI applications quickly using pre-trained FMs via an application programming interface (API). Amazon Bedrock makes FMs accessible from Amazon, as well as leading AI startups such as AI21Labs, Anthropic, Cohere, and Stability AI. These models allow developers to easily select and train the AI tool for their task, such as text generation, images, audio, and synthetic data creation in response to prompts.
Here are three things you may not know about the Amazon Bedrock service, outlining what might be possible as generative AI becomes more widely accessible:
1.Choose from a selection of machine learning foundation models
AI is a powerful resource that can be used across a variety of businesses to problem-solve or improve services. But each application will need to utilise different foundation models depending on the purpose. This is where Amazon Bedrock’s API access to multiple LLMs comes in.
Stable Diffusion, created by Stability AI, allows for the generation of unique, realistic, and high-quality images, art, logos, and designs on a large scale, whereas Jurassic-2, developed by AI21 Labs, provides multilingual text generation in Spanish, French, German, Portuguese, Italian and Dutch for business applications.Services from AWS combine decades of machine learning experience across Amazon with breakthrough technologies poised to transform just about everything.
Amazon Bedrock also provides exclusive access to Amazon’s own FMs – such as Amazon Titan. Amazon Titan FMs are pretrained on large datasets, making them powerful, general-purpose models. Amazon Titan can perform tasks such as text summarisation, text generation (for example, creating a blog post), classification, open-ended Q&A, and support responsible use of AI by reducing inappropriate or harmful content.
2.Customise models with your own training data
AI tools, and the FMs required to use them, are only as good as their data. For an improved experience, AWS users can easily and privately customise Amazon Bedrock FMs with their own proprietary data.
These customised FMs can create a unique customer experience, embodying the company’s voice, style, and services across a wide variety of consumer industries, like banking, travel, and healthcare. For example, a business that needs to auto-generate a daily report for internal circulation using all the relevant activity data can customise the model with proprietary data, including past reports, so that the FM learns how these reports should read and what datasets to report on.
Customised FMs have multiple exciting applications across different industries. Take healthcare, for example. Generative AI could lead to faster drug discovery and more accurate diagnoses. Health technology company Philips has already worked with AWS to develop generative AI applications to improve clinical workflows and diagnostics with imaging technology.From delivering medicine with drones to analysing fish with 3D cameras, Amazon Web Services customers use artificial intelligence data to tackle some of the world’s biggest challenges.
Manufacturers can also benefit from access to Amazon’s ML models, by analysing vast amounts of internet of things (IoT) telemetry to drive predictive maintenance, reduce downtime on production lines, generate new plans and designs for products that are stronger, have a lower impact on the environment, and a lower cost through generative design.
Security has always been a priority for Amazon, and Steve highlights the strong security controls AWS uses when handling sensitive information to train FMs. “Amazon Bedrock provides bar-raising security controls. Customers pass their data on AWS to Bedrock exclusively through the AWS network, and never via public internet. Privacy should be the cornerstone of any enterprise-ready generative AI technology, to ensure a customer’s private data and model customisations do not benefit other companies, including competitors.”
3.Lowering the entry requirements for AI development
“We're making these foundation models available for people to consume, rather than having to build them for themselves – saving months of potential effort, plus additional development and training costs” explains Steve. “The idea behind Amazon Bedrock is to provide an easy-to-consume, off-the-shelf way of accessing these powerful models and AI tools without the need for any data science knowledge.”
"At AWS, we have played a key role in democratising AI and ML, making it accessible to anyone who wants to use it, including more than 100,000 customers of all sizes and industries"Swami Sivasubramanian VP, Database, Analytics and ML at AWS
AWS believes in levelling the playing field when it comes to using AI, creating accessibility for startups and businesses to supercharge their work by removing cost and resource barriers. In 2017 the company launched Amazon SageMaker, which helps to open the door to AI and ML tools by providing a fully managed service that empowers everyday developers and scientists to use ML – without any previous experience. AWS also delivers custom-built chips and processors to give customers better price-to-performance for their applications.
“Since day one,” says Steve, “AWS has been focused on making ML accessible to everyone. Our approach to generative AI is to invest and innovate, to take this technology out of the realm of research, and make it available to customers of any size and developers of all skill levels.”