Amazon AWS Announces Full Launch of Amazon Bedrock and Series of AI Products and Updates

AWS announced the launch of five generative AI innovations to enable organizations of all sizes to build new generative AI applications, improve employee productivity, and transform businesses.

Swami Sivasubramanian, Vice President of Data and AI at AWS, said, "Over the past year, the increase in data, the availability of scalable computing, and advancements in machine learning have led to a surge of interest in generative AI, sparking new ideas that have the potential to completely transform industries and reimagine the way we work." He also noted, "Today's announcement is a significant milestone in bringing generative AI to every enterprise, from startups to large corporations, and to every employee, from developers to data analysts. Through powerful innovations, AWS provides customers with greater security, choice, and performance, while also helping them align their data strategies with the entire organization to fully unleash the transformative potential of generative AI."

1. Amazon Bedrock now officially launched to help more customers build and scale generative AI applications

Amazon Bedrock is a fully managed service that provides foundational models (FMs) from leading AI companies through a single application programming interface (API). FMs are very large machine learning (ML) models that have been pre-trained on massive amounts of data. The flexibility of FMs makes them suitable for a wide range of use cases, from search to content creation to drug discovery. However, many enterprises face challenges when adopting generative AI technology. Firstly, they need a simple way to discover and access high-performance FMs that excel and are best suited for their purposes. Secondly, customers want seamless application integration without the need to manage large infrastructure clusters or incur significant costs. Lastly, customers want to easily use the foundational FMs and build differentiated applications with their data. As customer-specific data for customization is highly valuable, it must be fully protected, secure, and private throughout the process, and customers also want control over how data is shared and used.

With the comprehensive capabilities of Amazon Bedrock, customers can easily experiment with various top-tier FMs and customize them with their proprietary data. Additionally, Amazon Bedrock offers unique features such as creating managed agents capable of performing complex business tasks, from booking travel and processing insurance claims to creating advertising campaigns and managing inventory, without the need for any coding. As Amazon Bedrock is serverless, customers do not need to manage any infrastructure and can securely integrate and deploy generative AI capabilities into their applications using AWS services they are already familiar with. Amazon Bedrock takes into account security and privacy, enabling customers to easily protect sensitive data.

2. Amazon Titan Embeddings now officially launched

Amazon Titan FM is a collection of models created and pre-trained by AWS on large datasets, making it a powerful general-purpose capability supporting various use cases. As the first model made available to customers, Amazon Titan Embeddings is a large language model (LLM) that converts text into numerical representations called embeddings, supporting search, personalization, and retrieval-augmented generation (RAG) use cases. Now, the next obvious question is, why would I do this?

FMs are great for various tasks, but they can only answer questions based on the training data and context provided, limiting their effectiveness in situations requiring timely knowledge or proprietary data. To enhance the response of FMs with additional data, many organizations turn to RAG, a technique that connects FMs to knowledge sources that can be referenced to enhance their responses. However, deploying RAG requires a significant amount of data and deep ML expertise, making it challenging for many organizations to achieve. This is where Amazon Titan Embeddings comes in.

Amazon Titan Embeddings makes it easier for customers to extend the capabilities of any FM by starting with RAG using their proprietary data. Amazon Titan Embeddings supports over 25 languages and context lengths of up to 8000 tokens (the longer the context length, the better the model can understand the conversation or text and generate accurate responses), making it ideal for use with words, phrases, or entire documents based on customers' use cases.

3. Meta's Llama 2 to be launched in the coming weeks

No single model fits all use cases, and to harness the value of generative AI, customers need access to a variety of models to discover the most effective ones for their own needs. That's why Amazon Bedrock enables customers to easily find and test a range of leading FMs, including those from AI21 Labs, Anthropic, Cohere, Stability AI, Amazon, and Meta, which will be launched in the coming weeks.

Amazon Bedrock is the first fully managed generative AI service that provides Meta's next-generation LLM, Llama 2, to users through a managed API. The Llama 2 model has significant improvements over the original Llama model, including training on 40% more data and a longer context length of 4,000 tokens to handle larger documents. Optimized for fast response on AWS infrastructure, the Llama 2 model available through Amazon Bedrock is particularly suitable for dialogue use cases.

4. New Amazon CodeWhisperer feature to allow customers to securely customize CodeWhisperer suggestions using their private code repositories, unlocking a new level of developer productivity

Amazon CodeWhisperer is an AI-driven coding assistant trained on billions of lines of Amazon and publicly available code to enhance developer productivity. Developers often use CodeWhisperer in their day-to-day work, but sometimes they need to incorporate their organization's internal private code repositories (such as internal APIs, libraries, packages, and classes) into their applications, which are not included in CodeWhisperer's training data. However, working with internal code can be challenging as documentation may be limited, and there may be no public resources or forums for developers to seek help.

The new customization capability of Amazon CodeWhisperer will enable the generation of AI-driven coding suggestions customized by securely leveraging customers' internal code repositories and resources, unlocking its full potential. Developers save time by improving the relevance of code suggestions. Administrators first connect to their private code repositories, such as GitLab or Amazon Simple Storage Service (Amazon S3), and schedule a task to create their customizations. This feature is built on enterprise-grade security and privacy, fully protecting customizations, and ensuring that the FM powering CodeWhisperer is not trained on customizations, thus safeguarding customers' valuable intellectual property. This customization feature will be available as part of the new enterprise version of CodeWhisperer for preview in the coming weeks.

5. New generative BI authoring capability in Amazon QuickSight to help business analysts easily create and customize visualizations using natural language commands

Amazon QuickSight is a cloud-native, unified business intelligence (BI) service that offers interactive dashboards, paginated reports, and embedded analytics, and provides natural language querying through QuickSight Q, ensuring every user in the organization can access the desired insights in their preferred format. Business analysts often spend a significant amount of time using BI tools to explore different data sources, add calculations, and create and refine visualizations before delivering them in dashboards for business stakeholders. To create a chart, analysts must first find the right data source, identify data fields, set filters, and make necessary customizations to ensure visually appealing visualizations. If the visual effect requires new calculations (e.g., annual sales), analysts must identify the necessary reference data and then create, validate, and add the visual effect to the report. Organizations benefit by reducing the time business analysts spend manually creating and adjusting charts and calculations.

The new generative BI authoring capability extends QuickSight Q's natural language querying from answering structured questions (e.g., "What are the top ten products in California sales?") to helping analysts quickly create customizable visualizations from fragments of questions (e.g., "top ten products") by asking follow-up questions to clarify the intent of the query, refine the visual effect, and complete complex calculations. Business analysts simply describe the desired outcome, and QuickSight generates visually appealing visualizations that can be easily added to dashboards or reports with a click.