Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
Amazon Web Services (AWS), announced today that it is expanding its generative AI services in a bid to make the technology more available to organizations in the cloud.
Among the new AWS cloud AI services is Amazon Bedrock, which is launching in preview as a set of foundation model AI services. The initial set of foundation models supported by the service include ones from AI21, Anthropic, and Stability AI, as well as a set of new models developed by AWS known collectively as Amazon Titan.
In addition, AWS is also announcing the general availability of Amazon EC2 Inf2 cloud instances powered by the company’s own AWS Inferentia2 chips, which provide high performance for AI.
Rounding out the updates, the Amazon CodeWhisperer generative AI service for code development is now generally available, with AWS making it free for all individual developers.
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
“One of our key goals behind all of these announcements and launches is to democratize the use of generative AI,” Bratin Saha, VP of ML and AI services at Amazon, told VentureBeat.
The expanded AI push from AWS comes as its cloud rivals — including Microsoft Azure and Google Cloud — continue to roll out their own respective sets of services, and as organizations of all sizes look to benefit from AI.
Amazon Bedrock lays a new foundation for AI in the cloud
With the Amazon Bedrock service, the goal is provide users with a set of foundation models that they can choose from.
The models can then be customized with additional training in AWS to suit whatever a user needs. Saha emphasized that a key benefit is the fact that the Bedrock service is integrated with the rest of the AWS cloud platform. That means organizations will have easier access to data they stored in Amazon S3 object storage services, as well as being able to benefit from AWS access control and governance policies.
“The fact that customers will be able to use foundation models with the AWS enterprise security and privacy guarantees, we think, makes it much easier for using these models at scale,” Saha said. “Customers will be able to use Amazon Bedrock in the same environment and with the same AWS services that they’re already comfortable with using.”
AWS enters the foundation model arena with Titan
As part of the Bedrock announcement, AWS is also making its own Titan model available.
Saha explained that AWS built Titan on its own to provide an alternative model for organizations. At launch there are two different flavors of Titans, one being a text model for content generation, the other being what Saha referred to as an embedding model. He explained that the embedding models create vector embeddings and can be used for things like creating highly efficient search capabilities.
The size and scale of the Titan models, in terms of the number of parameters, is not something Saha was able to comment on. Parameters are sometimes used as a way to measure the size of a model; for example, GPT-3 is 175 billion parameters. Saha commented that in his view, parameters are not necessarily a good indicator of how well a model will perform.
“We have been building Titan working backward from customer use cases,” Saha said.
For Titan, as well as the other foundation models in Amazon Bedrock, Saha emphasized that responsible and explainable AI is a critical component.
“In general, responsible AI is, no pun intended, a bedrock for everything we do,” Saha said.
For Amazon Bedrock specifically, he said that AWS is making sure that the datasets are being filtered for inappropriate content. For example, for the generated output there are filters to make sure that the output from the modules is appropriate.
“We have a very comprehensive responsible AI program that is already being used for our current AI services and, for Bedrock, we are just enhancing it,” Saha said.
CodeWhisperer brings generative AI to developers for free
Amazon CodeWhisperer was first announced by AWS in September 2022 and has been in preview until today.
CodeWhisperer is a competitive alternative to GitHub Copilot, which is powered by OpenAI’s Codex large language model. In contrast, CodeWhisperer uses a model that AWS has built on its own.
Saha said that since CodeWhisperer was first announced, AWS has improved the service that helps developers use generative AI to write code. Among the improvements is lower latency and, perhaps more importantly, a reference tracker. If the code that CodeWhisperer generates is similar to other code, the reference tracker system will provide the proper attribution. Providing attribution for code is critical as it makes it easier and safer for generated code to be used in enterprise settings.
Going a step further, CodeWhisperer also provides enhanced security as it validates all generated code and checks it for potential vulnerabilities.
“We think by making CodeWhisperer free for every developer, we can really democratize access and make developers a lot more productive,” he said.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.