Amazon Web Services Inc. announced today during AWS Summit New York that it’s expanding its fully managed artificial intelligence foundational model service Bedrock to include a variety of new models from Cohere, Anthropic and Stability AI, as well as the capacity for developers to deploy their own automated agents to execute complex tasks.

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

Generative AI has opened up a wide variety of application use cases across numerous industries where models can be used to plump company data resources to generate insights, write emails, do research, produce graphs, generate artwork and more.

“Generative AI has the potential to transform every application, business, and industry. Advancements across data processing, compute and machine learning are expediting the shift from experimentation to deployment for AWS customers of all sizes,” said Swami Sivasubramanian (pictured), vice president of database, analytics and machine learning at AWS.

Cohere Inc. is the newest foundational model provider coming to Bedrock. It delivers a text-based generation model called Command designed for enterprise-level business applications. Command is trained to follow user commands and is capable of summarization, copywriting, dialogue, extraction and question-answering. Cohere also produces a text understanding model called Embed, which provides search and classification tasks for more than 100 languages, enabling applications that can quickly search context and meaning within text.

Bedrock is also receiving Claude 2, Anthropic’s most recent upgrade to its Claude chatbot. It can ingest up to 100,000 tokens for each text prompt, which will allow it to ingest about 75,000 words. That means users will be able to submit extremely long documents and also allow the AI to write longer documents in response, on the order of a few thousand tokens.

Stability AI Ltd., the AI startup behind the text-to-image generative AI model Stable Diffusion, is bringing its most recent iteration of generative image models to Bedrock with Stable Diffusion XL 1.0. Using natural language text inputs, users can tell SDXL 1.0 what they want it to produce and it will generate vivid and beautiful images, often with next-level photorealism. The newest model is even more capable than previous generations, bringing superior quality and enhanced capabilities in image composition and realism. according to the company.

“Stability AI thrives on innovation, and our foundation model Stable Diffusion on Amazon Bedrock reflects the most pioneering developments across generative AI,” Emad Mostaque, founder and chief executive of Stability AI. “Offering the latest release of our foundation model SDXL 1.0, with improved image and composition detail over its predecessor, gives Amazon Bedrock customers access to cutting-edge resources and advances our goal to activate humanity’s potential with AI.”

SDXL 1.0 is also being released today on Amazon SageMaker JumpStart, a machine learning hub for developers that delivers access to algorithms, models and solutions.

Agents for Amazon Bedrock automate complex AI tasks

Although foundational models are capable of understanding conversational language prompts and producing text or images in response – for tasks such as summarization or question answering – they cannot complete complex tasks that application calls or interaction with third-party systems.

To make this happen, Amazon Bedrock is introducing the fully managed capability for developers to create generative AI applications that can call upon “agents,” which can break down tasks into multiple steps that can accomplish a series of tasks on demand.

Without any manual coding, Agents for Amazon Bedrock will automatically break down tasks and generate an orchestration plan based on the needs of the developer. The agent then securely connects to company data through application programming interfaces.

For example, an agent could be used to have a generative AI model accomplish the task of setting up a time off request for an employee or booking a plane flight. All of this would require the model to understand what the user asked for and then do a step-by-step series of tasks to complete the needed setup. In each case, there might even be follow-up questions that need to be answered by the employee or customer that the AI model may need clarification on, or updates for that may need up-to-date information as well, requiring connectivity or further tasks that the agent needs to handle.

Bedrock Agents manage all of this behind the scenes securely by provisioning all of the resources needed by doing the API calls, getting the resources and handling the system integration for developers so they can do the application work.

Photo: AWS/livestream