Amazon Web Services Inc. today rolled out five new generative AI offerings across its ecosystem that will help business customers build AI-enabled applications with better security with accessibility to models using their own data.

These new services include the general availability of Amazon Bedrock, availability for Meta Platforms Inc.’s Llama 2, AI embeddings from Amazon Titan, new coding capabilities for Amazon CodeWhisperer and generative AI enhancements for business intelligence with QuickSight.

“Over the last year, the proliferation of data, access to scalable compute, and advancements in machine learning have led to a surge of interest in generative AI, sparking new ideas that could transform entire industries and reimagine how work gets done,” said Swami Sivasubramanian, vice president of data and AI at AWS. “Today’s announcement is a major milestone that puts generative AI at the fingertips of every business, from startups to enterprises, and every employee, from developers to data analysts.”

Amazon Bedrock is the company’s fully managed service for generative AI that provides access to foundational models. Customers can use the service to discover, train and fine-tune their models with their own proprietary data on Amazon’s high-performance infrastructure in a secure environment all without the need to spin up or manage anything.

Now in general availability, customers will have immediate access to high-performing models from leading AI companies such as AI21 Labs, Anthropic, Cohere Inc., Meta Platforms Inc., Stability AI Ltd. and Amazon’s custom models as well. The service also provides a large number of capabilities that customers can use to build their own AI-enabled applications that can respond conversationally to customers, summarize documents, generate images, and provide AI driven search.

Amazon boasts that Bedrock is the first fully managed service to offer Meta’s Llama 2 large language model through an application programming interface. In the next few weeks, developers will be able to build AI-enabled applications using Llama 2, optimized for use on AWS infrastructure, through Bedrock, based on the 13 billion- and 70 billion-parameter models.

Titan Embeddings is now generally available, allowing customers to quickly create AI-enabled applications based on large datasets. Amazon Titan foundational models are a family of models that are based on large datasets that convert text to numerical representations called embeddings, these representations can then be used to augment contextual search based on semantic meaning. This can be used to enhance AI-powered searches, provide better personalization and other use cases.

Since building embeddings models requires vast amounts of data and significant machine learning expertise, it can be difficult for many businesses to implement the capability. Now with Titan Embeddings enterprise customers can have it at their fingertips with a managed service. The capability on AWS is available in more than 25 languages and has a context length of 8,192 tokens, which means it can handle anything from single words to extremely long documents.

Amazon CodeWhisperer acts as an AI-powered coding companion for developers that works with them by suggesting code snippets, rewriting code and explaining code. Its model has been trained on billions of lines of publicly available source code, which enables it to provide these capabilities to developers. Amazon has upgraded it so that enterprise customers can customize it with private code from their internal codebase so it can provide recommendations based on their own unique requirements.

Before this update, a developer might use the tool to help them write generic code, but CodeWhisperer would not have been aware of the company’s specific internal needs or coding practices. A developer might have still had to spend hours of time working through previously written code and inspecting it to ensure that it follows the company’s practices.

With the new customization capability, CodeWhisperer will be able to build on already existing code and maintain already great quality code along with developers by bringing their tooling together consistently. Amazon said that it will do so in a way that will not leak confidential information and will not store or log any customer information from the customer’s development environment.

“We are equipping our 16,000-plus engineering organization with Amazon CodeWhisperer to build and deliver industry applications faster and more securely in a responsible way,” said Pandurang Kamat, chief technology officer at enterprise modernization company Persistent Systems Ltd. “Several teams have started to leverage CodeWhisperer’s new customization capability to help maximize the benefits from generative AI-powered code suggestions, and we are already seeing great results.”

A recent study done in collaboration between Persistent and AWS found that developers were able to complete coding tasks up to 28% faster on average with the new capability.

Business users received new generative AI capabilities with Amazon QuickSight, a unified business intelligence service build on the cloud that provides interactive dashboards, reports and embedded analytics. QuickSight has natural language querying capabilities using QuickSight Q, this allows any user to get insights just by typing in a structured question.

Amazon is introducing new capabilities to QuickSight Q that extends its natural language capabilities so that analysts can be looser with their language when prompting analytics engine for information and insights. Before the update, it required analysts to write well-structured questions such as “What are the top 10 products in Arizona?” But with the power of generative AI, it’s now more capable of handling far more complex ideas.

Now, analysts can simply describe the outcome they want to get their desired visualizations and ask follow-up questions based on the reports generated to refine complex calculations. Once completed, the answers and charts can be quickly added to a dashboard or report with a single click.

For example, now an analyst could ask the AI to create a visualization for a “monthly trend for citrus sales in 2022 and 2023,” and it will select information in a chart format such as a line chart or bar chart that makes the most sense for that request. If the analyst prefers a different kind of chat, a followup request could change it to a spreadsheet.

The new generative AI capability will also allow QuickSight Q to offer related questions to analysts when there are ambiguous cases if multiple data fields match a query.

Image: AWS