This year, taking advantage of the latest trends requires firm feet on the ground, while working with a good partner to achieve necessary goals….
At AWS re:Invent this week, Amazon Web Services (AWS) unveiled major upgrades to Amazon Bedrock, its fully managed platform for building and deploying generative AI applications. The latest innovations vastly expand model choice, deliver powerful new capabilities, and make it even easier for organisations to harness generative AI.
More Model Options Than Ever Before
AWS has added models from AI leaders Anthropic, Cohere, Meta and Stability AI to Amazon Bedrock’s already robust catalogue. This includes Anthropic’s Claude 2.1, Cohere’s multilingual offerings, Meta’s Llama 2, and Stability AI’s Stable Diffusion XL. AWS has also expanded its own family of Amazon Titan models tailored for Bedrock with the new Image Generator for creating synthetic media and Multimodal Embeddings for semantic search.
With the broadest selection of foundation models available anywhere, developers now have unmatched flexibility to evaluate, combine and optimise models to suit their specific use case.
Revolutionary Tools to Customise and Refine Models
Innovations like fine-tuning and knowledge bases help developers tap into the power of their proprietary data to make models more accurate and responses more relevant. Support for retrieval augmented generation connects models directly to private databases and documents to generate customised, up-to-date responses.
Meanwhile, managed fine-tuning enables users to further train models on labelled data specific to their business and industry terminology. Models from Cohere, Meta and Amazon Titan can now be fine-tuned with just a few clicks.
More Intelligent Agents That Integrate and Automate
The newly released Agents capability allows models to execute complex, multi-step tasks by programmatically interacting with internal systems and external APIs. Developers simply define the task in natural language, and Agents handle the orchestration automatically behind the scenes.
Agents can also take advantage of knowledge bases to inject context and pull real-time data into responses. This makes it easy to build sophisticated applications that blend transactional capabilities with generative AI.
Guardrails Drive Responsible Innovation
As generative models grow more advanced, guardrails help set the boundaries for safe, helpful and constructive interactions. Developers can now define custom filters, blocklists and thresholds to control the topics, language and content allowed across applications.
Guardrails enable brands to block unwanted responses, enforce policies consistently, and prevent harmful or biased output—all while encouraging creativity and maintaining privacy.
The Power of Generative AI, Made Easier Than Ever
With breakthrough expansions to Amazon Bedrock’s features and foundation models, AWS is removing the barriers standing between enterprises and the responsible application of leading-edge generative AI. Bedrock makes it simple to build next-generation experiences with the latest techniques in machine learning—unleashing innovation while keeping customer data secure.