Amazon Bedrock opens the door to two new large language models hosted on AWS. Plus, AWS added generative AI to QuickSight and CodeWhisperer.
Amazon has added generative artificial intelligence models into Amazon Web Services.
With Amazon Bedrock, customers can build and scale generative AI-based applications through an API. From there, the libraries of Amazon Titan FMs, which include two large language models, can be accessed, along with other models from AI21 Labs, Anthropic, Cohere, Meta, Stability AI and Amazon.
Jump to:
Amazon Bedrock is a service that provides access to generative AI on the cloud. Technically, Amazon Bedrock is a library of foundation models, all of which provide similar content-generating abilities. Since Amazon Bedrock’s models are hosted by AWS, existing clients can access them through their usual channels and not have to worry about hosting the model’s large amounts of data on their own infrastructure. However, enterprise clients will be able to feed data sets of their choice into the models.
Perhaps mindful of worries about security, Amazon reassured people that data fed into a training model in this way would not be used to train future iterations back at Amazon HQ.
Customers will also be able to use familiar AWS controls for the new AI. It connects to the cloud-based machine learning platform SageMaker ML, which includes an environment in which to test different models and a feature for managing foundation models at scale.
SEE: Looking to get started with ChatGPT?
On September 28, 2023, AWS announced that Llama 2, Meta’s current-generation large language model, will be available in Amazon Bedrock through a managed API; it is expected to be accessible in mid- to late October. Amazon Bedrock subscribers will be able to choose between the 13B or 70B parameter Llama 2 models.
AWS customers can sign up for an Amazon Bedrock subscription now. Amazon Bedrock pricing varies based on specific services; the full list of services is on the pricing page.
While Amazon Bedrock is the service, Amazon Titan is the content. Amazon Titan plans to offer two AI models: one that creates text and one that improves searches and personalizations. In addition to creating new text such as blog posts or emails, Amazon Titan can sort items into different classifications, hold open-ended conversations and extract specific information from chunks of text.
On September 28, AWS announced the first available Amazon Titan model called Titan Embeddings, which enhances searches by tailoring the results to more relevant and contextual responses. Titan Embeddings is a large language model trained for retrieval-augmented generation, a model-customization technique that allows organizations to include timely knowledge or proprietary data in searches. A model similar to this one is already at work in Amazon’s product search.
Once organizations have the AI hooked up to their existing AWS, Amazon said they can customize a model by feeding it just 20 examples of the kind of task that particular organization might want the AI to complete.
“We predict that most businesses’ long-term approach will be to customize large language models to their needs and data instead of out-of-the-box model usage,” a Slalom representative told Amazon.
Accenture, Deloitte, Infosys and Slalom were early integration partners.
Coda was one of the organizations using Amazon Bedrock in preview. “Since all our data is already on AWS, we are able to quickly incorporate generative AI using Bedrock, with all the security and privacy we need to protect our data built-in,” said Shishir Mehrotra, co-founder and CEO of Coda.
Another possible AI use case specifically for developers comes from CodeWhisperer, an AI-based coding assistant. In April 2023, Amazon announced CodeWhisperer now has a free tier. The no-cost Individual tier provides access to automated code-writing that lives inside a developer’s existing integrated development environment.
CodeWhisperer works well for “creating code for routine or time-consuming, undifferentiated tasks, working with unfamiliar APIs or SDKs, making correct and effective use of AWS APIs, and other common coding scenarios such as reading and writing files, image processing, writing unit tests,” Amazon wrote in a press release. The Professional tier, which is available now, offers more features, such as single sign-on and identity and access management Identity Center integration and higher limits on security scanning.
On September 28, 2023, AWS expanded CodeWhisperer with a generative AI recommendation capability to help developers problem-solve in their internal, private APIs, libraries, packages and classes. Developers can provide a GitHub or Amazon S3 link to their private repository from CodeWhisperer. The customized recommendation capability is expected to be available in preview in the upcoming Amazon CodeWhisperer Enterprise tier “soon,” AWS said in a press release.
AWS and Amazon Titan do not use internal customizations for training or store or log any customer content from the Amazon CodeWhisperer Professional or Enterprise tiers, AWS said.
Business intelligence analysts can leverage generative AI to create visuals in Amazon QuickSight with Amazon QuickSight Q, which is now available as a free trial. QuickSight Q can answer questions or create charts and graphs based on natural-language queries. The visuals can then be added to reports or dashboards created in QuickSight.
Amazon Bedrock’s major competitors are the enterprise subscription for OpenAI’s ChatGPT and Google’s PaLM 2 model, which is available for enterprise customers through Google’s Vertex AI platform.
