During the keynote address at the AWS re:Invent 2024, CEO Matt Garman introduced an array of upgrades and new features.
AWS CEO Matt Garman made a slew of announcements about the company’s AI offerings on Tuesday, December 3. At Amazon’s flagship cloud event, AWS re:Invent 2024, Garman reaffirmed the company’s longstanding commitment to innovation, with customers at its core. During the keynote, Garman announced upgrades to Amazon Bedrock, its platform designed to help companies build generative AI tools; the next generation of Amazon SageMaker; upgrades to Amazon Q Business; new capabilities for Amazon Q Developer; and more.
Below is a look at some key announcements from the high-stakes keynote by Matt Garman.
Amazon Bedrock gets new AI safeguards and more
During his keynote, Garman introduced some enhancements to Amazon Bedrock, essentially making it more safe and efficient. The platform now has factual accuracy with automated reasoning, which is groundbreaking as it is a safeguard against AI hallucinations. Garman said that the tool offers logically accurate outputs, which is crucial for healthcare and finance sectors. Factual accuracy and automated reasoning are integrated with Bedrock Guardrails, and it validates responses, offers audit trails, and streamlines AI outputs. Another key update is the multi-agent collaboration, which means Amazon Bedrock agents will now support complex workflows by managing specialised AI agents. These agents will be able to work across diverse data sources to automate the most complex tasks, like financial analysis. Amazon Bedrock also gets cost-effective model customisation, which is essentially its new model distillation feature that allows customers to create smaller, faster models with performance on par with larger models. And it does this without the need of machine learning expertise. Garman said that users can fine-tune models for specific use cases, reducing costs and latency, making them suitable for real-time applications.
Next-gen Amazon SageMaker
Launched in 2017, Amazon SageMaker is essentially a cloud-based machine-learning platform that lets developers create, train, and deploy ML models on the cloud. At the event, Garman introduced the next generation of Amazon SageMaker with a slew of features such as SageMaker Unified Studio, SageMaker Catalog, SageMaker LakeHouse, and Zero-ETL integrations. While the SageMaker Unified Studio offers a centralised environment for data and AI development, allowing seamless access to organisational data, SageMaker Catalog enhances security and compliance by managing access to data and AI models.
On the other hand, SageMaker Lakehouse offers unified access to diverse data sources, such as Amazon S3 and Redshift. Zero-ETL integrations let customers connect SaaS applications like Zendesk and SAP directly to SageMaker, eliminating the need for complex data pipelines. It needs to be noted that SageMaker Unified Studio is currently available in preview.
Software development made easy with Amazon Q Developer
At the re:Invent event, Amazon announced upgrades to Amazon Q Developer, a generative AI assistant that has been developed to enhance software development. The new updates aim to make life easier for developers by automating tasks like unit testing, documentation, and code reviews, tasks that are usually seen as tedious. Besides, Amazon Q Developer can now autonomously identify and generate unit tests. It also automates the creation and maintenance of accurate documentation, making it easier for developers to keep project details updated. The tool now offers automated feedback, enhancing code reviews. With the new operational feature, developers and operators can instantly resolve issues by analysing large amounts of AWS data, locating the root cause of an issue, and helping users through fixes.
The new Trainium3 chip
The cloud computing giant also introduced Trianium3, which is its next-generation AI chip that has been developed to facilitate generative AI workloads. The new chip is built on a 3-nanometre process and offers improved performance, density, and power efficiency. The new chip is expected to power UltraServers that are four times more efficient than their predecessors. The first instances powered by Trainium3 will be available by late 2025.
Amazon Nova models
During the keynote, Amazon CEO Andy Jassy announced Amazon Nova, its new generation of foundation models. These models come with the ability to process text, images, and video as prompts. Essentially, consumers can use Amazon Nova-powered Gen AI applications to understand videos, charts, and documents, or generate the same. The Amazon Nova models available in Amazon Bedrock include the Amazon Nova Micro, Nova Lite, Nova Pro, Nova Premier, Nova Canvas, and Nova Reel. Jassy said that all the Nova models are incredibly fast, cost-effective, and have been developed to be used with ease. The Amazon CEO also demonstrated the performance of the new Nova models across various benchmarks. While Nova Lite seems to excel at cost-efficient multitasking, Nova Pro strikes a balance between speed and accuracy for diverse tasks, and Nova Micro has been optimised for low-latency responses. The new Nova models are versatile and showcase excellence across text, visual reasoning, adaptive workflows, etc.
AWS has introduced numerous new features and models, and the list above is some of the most notable updates and announcements. By bringing generative AI to almost every aspect of its offering, AWS seems to be taking on Microsoft. With advancements like Trainium3 processors and going deeper with AI integration into its services, AWS is marching ahead within the generative AI race among enterprise solution providers. The Amazon company is also offering AI tools to migrate Windows apps to AWS.
Content retrieved from: https://indianexpress.com/article/technology/artificial-intelligence/aws-reinvent-2024-from-upgrades-to-amazon-bedrock-to-nova-foundation-models-key-takeaways-9705200/.
