On the second day of AWS re:Invent 2024, Dr Swami Sivasubramanian, vice president of AI and Data at AWS, announced an array of new advancements for Amazon Bedrock, the company’s platform that allows businesses to build generative AI applications. On the first day, we saw AWS CEO Matt Garman and Amazon CEO Andy Jassy introducing some new models and capabilities. On Thursday, December 5, Sivasubramanian showcased new model capabilities with a glimpse of how AWS is affecting change in their respective industries with generative AI.
The new upgrades aim to offer greater flexibility and control to build and deploy generative AI applications faster and more efficiently. All the announcements show AWS’s commitment to model choice and optimising how inference is done at scale. In his insightful keynote, Sivasubramanian asserted that AWS, with its groundbreaking technology, was not only shaping the present but also laying the groundwork for future innovations to take flight.
Here is a look at key moments from the keynote:
What’s new with Amazon Bedrock?
During his keynote speech, Dr Sivasubramanian introduced some new features for Amazon Bedrock. The latest updates include expanded model options, access to more than 100 specialised models through the Amazon Bedrock Marketplace, enhanced prompt management tools, and some new features for knowledge bases (a self-serve online library of information) and data automation. Dr Sivasubramanian said that these features aim to offer flexibility, scale inference, and maximise data use. While other capabilities are in preview, the Amazon Bedrock Marketplace is live. The AWS executive also said that models from Luma AI, Poolside, and Stability AI will soon be added to Amazon Bedrock.
New Amazon SageMaker AI capabilities
AWS’s service of building and deploying AI models—Amazon SageMaker—got four innovations aimed at making generative AI and machine learning development cost-efficient, faster, and easier to scale. The new innovations focus on helping companies start instantly with popular models, optimising their training processes, and integrating seamlessly with partner AI tools. The features are curated training recipes, flexible training plans, task governance, and integrated partner AI apps.
What these advancements mean for customers is that they will get faster and more affordable AI solutions. Businesses can now expect more personalised, efficient, and innovative experiences, like smarter chatbots, quicker recommendations, and improved automation in day-to-day tasks.
Amazon Bedrock Marketplace
One of the significant announcements on Day 2 was the new Amazon Bedrock Marketplace. This is a venue that offers access to over 100 popular and specialised AI models, including Mistral NeMo, Falcon RW, etc. Users can opt for models tailored to their needs, deploy them on scalable AWS infrastructure through fully managed endpoints, and securely integrate them using Bedrock’s APIs. It also includes guardrails, agents, and strong security and privacy protections. According to AWS, this marketplace simplifies model discovery, deployment, and integration.
SageMaker HyperPod gets new features
To address the growing demands of AI, AWS has announced some new features for SageMaker HyperPod. These included flexible training plans to streamline capacity reservations, saving weeks of training time and allowing work within budgets and timelines. On the other hand, task governance in SageMaker HyperPod automates the management and prioritisation of compute resources, maximising utility and completing high-priority tasks with efficiency. SageMaker also integrates AI apps from partners like Comet and Fiddler, essentially reducing time spent configuring third-party tools and speeding up the model development lifecycle. These innovations are aimed at enhancing resource efficiency, reducing development complexity, and improving AI deployment speed for customers.
Advanced Gen AI enhancements on Bedrock
Dr Sivasubramanian in his keynote, also introduced a suite of innovations that simplify and optimise generative AI development. While prompt caching assists with reusing context across API calls, intelligent prompt routing improves response quality and cost-efficiency by directing queries to the best-suited AI model. To address challenges related to Retrieval Augmented Generation (RAG), AWS introduced Kendra Gen AI Index, which integrates with over 40 enterprise data sources for accurate and engaging outputs.
On the other hand, Bedrock Knowledge Bases now enable structured data retrieval and the use of knowledge graphs through GraphRAG support, which allows for richer, more precise responses in Gen AI applications. AWS also introduced Bedrock Data Automation, which processes unstructured multimodal data to enhance Gen AI insights.
For safety and to ensure ethical use, AWS introduced Bedrock Guardrails that offer customisable safeguards and automated reasoning checks. Meanwhile, the new multimodal toxicity detection filters harmful image content.
The author is at AWS re:Invent 2024 in Las Vegas at the invitation of AWS.
Source: AWS re:Invent 2024: Swami Sivasubramanian unveils Bedrock Marketplace, SageMaker updates,
