AWS re:Invent 2024:
AI gets practical

Amazon Web Services (AWS) re:Invent 2024 this week marked a pivotal moment in the evolution of generative AI, as AWS unveiled a suite of tools and partnerships designed to make artificial intelligence more accessible, scalable, and practical for businesses of all sizes. The second day of the event, on Wednesday, saw massive updates to foundational platforms like Amazon Bedrock and startling new capabilities in Amazon SageMaker AI.

Central to the announcements was Amazon Bedrock, AWS’s fully managed service for building and scaling generative AI applications. Dr Swami Sivasubramanian, AWS VP of AI and data, captured the significance of the moment: “We are at a tipping point where generative AI is moving out of experimentation and into mission-critical applications. Businesses need tools that are not only powerful but also intuitive and cost-effective.”

Bedrock’s updates showcased this philosophy. Key among the new features were Intelligent Prompt Routing and prompt caching, designed to optimise cost and performance when running generative AI queries. 

“Prompt caching isn’t just about saving money – it’s about ensuring that businesses can scale AI solutions without sacrificing efficiency or accuracy,” said Sivasubramanian.

Bedrock Data Automation transforms unstructured, multimodal data into structured formats without requiring coding. This capability addresses a longstanding challenge for many businesses: the vast majority of their data is not AI-ready. 

“Data Automation is a game-changer,” said Sivasubramanian. “It eliminates one of the biggest barriers to leveraging data effectively, enabling organizations to tap into the full potential of their information.”

The addition of GraphRAG to Bedrock’s Knowledge Bases further expanded its utility, allowing businesses to integrate structured and graph data into generative AI workflows.

 “This feature opens up entirely new possibilities for industries like healthcare and finance, where nuanced understanding and contextual data are crucial.”

AWS’s partnerships also took center stage, with the announcement that Luma AI and poolside would soon bring their models to Amazon Bedrock. Luma AI’s Ray 2 Video Model, for instance, promises to revolutionise video production by enabling users to generate high-quality, production-ready videos from text or image prompts.

“These models aren’t just tools for specialists,” said Sivasubramanian. “They’re enablers of creativity and innovation across a wide range of industries.”

poolside, a frontier AI company specialising in software engineering, will make its generative AI assistant and foundational models available on Bedrock. These models are optimised for tasks like code generation and testing, powered by AWS’s Trainium chips, which offer unparalleled efficiency in AI inference. 

“Our collaboration with poolside reflects our commitment to empowering developers with the best tools for every stage of their workflow,” said Sivasubramanian.

AWS also unveiled significant updates to Amazon SageMaker AI, its flagship platform for building, training, and deploying machine learning models. Among the enhancements were three new features for SageMaker HyperPod, including Flexible Training Plans, which simplify model training workflows; tools to maximize compute resource utilisation, cutting costs by up to 40%; and the integration of partner applications such as Comet and Deepchecks directly into SageMaker. “Training large models is no small feat,” said Sivasubramanian. “With HyperPod, we’re reducing complexity and enabling breakthroughs that were previously out of reach.”

One of the standout features of HyperPod is Task Governance, which dynamically allocates compute resources across multiple projects, ensuring maximum utilization. “In a world where compute resources are both finite and expensive, Task Governance is about making every second of computation count,” said Sivasubramanian.

Beyond infrastructure and tools, AWS demonstrated a commitment to democratising AI and empowering a broader audience, with the launch of Amazon Canvas, a platform for generating studio-quality images and videos using text prompts. 

“Canvas isn’t just for developers,” said Sivasubramanian. “It’s for marketers, designers, and creators who want to bring their ideas to life without needing deep technical expertise.”

Sustainability and social impact were also key themes, with AWS announcing a $100-million commitment to providing underserved students with skills in AI and cloud computing. “The future of AI isn’t just about technology,” said Sivasubramanian. “It’s about people. By investing in education, we’re ensuring the next generation has the tools to thrive in an AI-driven world.”

These innovations represent more than technical milestones — they signal a vision for the future of AI. Sivasubramanian summed it up: “AI is no longer a tool for a select few. It’s a transformative force for everyone, whether they are developers, businesses, or creators. Our role at AWS is to make sure that transformation is as seamless, impactful, and accessible as possible.”

* Arthur Goldstuck is CEO of World Wide Worx and editor-in-chief of Gadget.co.za. Follow him on Bluesky on @art2gee.bsky.social.