Companies like Microsoft – in partnership with startup OpenAI – Google and others have grabbed a lot of the headlines about generative AI for the past year or so. At one point, ChatGPT was the fastest-growing internet application of all time, and Google closely followed the release of that tool with its own chatbot, Bard.

As these IT vendors scatter ChatGPT, Bard and similar tools throughout their product portfolios, enterprises and consumers alike are eagerly adopting them.

Unsurprisingly, Amazon Web Services’ (AWS) New York Summit last week was all about generative AI. But the focus was more on the tools and services the public cloud titan can offer organizations to build their own AI applications rather than what AWS itself has created.

In a blog post released as the show got underway, Swami Sivasubramanian, vice president of database, analytics, and machine learning at AWS, leaned into that approach, writing about the ways generative AI will help businesses in such areas as customer experience, employee productivity, business operations, and creative content.

“The key is to ensure that you actually pick the right AI-enabled tools and couple them with the right level of human judgment and expertise,” Sivasubramanian wrote, adding that “generative AI has the potential to revolutionize our lives, whether at home, school, or work. With these tools Amazon and our customers are building, we’ll all be able to spend more time on what we are best at, and less time on the routine work.”

During the event, AWS unveiled new and enhanced tools for creating models with its Bedrock fully managed foundation model (FM) service, getting better insights insight from data, offering more training and certifications, and adding more much-needed compute power to handle the demands of large-language models (LLMs). The company also showed how its expansive services portfolio can help healthcare organizations integrate generative AI into their applications.

The Need for Responsible AI

And like Microsoft, Google, OpenAI, and others, AWS said it is working to put the safeguards in place to ensure the generative AI it enables doesn’t become the societal nightmare some worry about.

“We continually test and assess our products and services to define, measure, and mitigate concerns about accuracy, fairness, intellectual property, appropriate use, toxicity, and privacy,” wrote Peter Hallinan, head of responsible AI at AWS. “And while we don’t have all of the answers today, we are working alongside others to develop new approaches and solutions to address these emerging challenges.”

Sivasubramanian said the industry has reached “a tipping point” with AI, fueled by improved technology, that massive amount of data being generated, the growth of highly scalable compute and storage capabilities–particularly in the cloud–and innovations around machine learning technologies. What once was the playground of scientists is now being applied by businesses and consumers.

Bulking Up the Bedrock for AI

FMs are the large machine learning models driving generative AI, trained on massive amounts of data. At the summit, AWS put a lot of focus on its Bedrock service, which was announced in April and is now in preview. AWS added FMs from four-year-old startup Cohere as well as Anthropic and Stability AI, both of which launched in 2021, to broaden the options for models that enterprises can use to build their applications.

AWS also announced the preview of agents for Bedrock. Developers can create fully managed agents which enable generative AI applications to run complex tasks that require making API calls to enterprise systems. Agents for Bedrock make the complicated process of making AI applications simpler, requiring only a few clicks and don’t require developers to hand-code anything. The agent connects to an enterprise’s data via an API and converts the data into a format that can be read by machine learning systems.

The cloud giant said its Entity Resolution analytics service is generally available, automatically matching and linking related records housed in different parts of an enterprise, from applications to channels to data stores. It also added new LLM capabilities to its Quicksite unified business intelligence service, bringing generative BI capabilities to Bedrock.

In addition, AWS announced a preview release of a vector engine to improve the search performance in OpenSearch Serverless.

Building the Infrastructure With NVIDIA

The company also leveraged its decade-plus-long partnership with GPU maker NVIDIA to add more compute power to run and scale AI and HPC workloads. GPU accelerators are foundational to building the high-performance and highly scalable infrastructure necessary for the compute- and data-hungry AI applications.

AWS and NVIDIA in April said they were working together to build infrastructure for training massive LLMs and building generative AI applications. Last week, they unveiled EC2 P5 instances, powered by NVIDIA’s latest T100 Tensor Core GPUs.

The new instances will deliver a six-fold reduction in the amount of time needed to train LLMs and computer vision models over previous GPU-powered instances. The P5 instances include eight Nvidia GPUs, third-generation AMD Epyc server CPUs, 2TB of memory, and 30TB of local NVMe storage. All that adds up to training times that take hours rather than days, according to Channy Yun, principal developer advocate for AWS.