1. Powerful New Capabilities for Amazon Bedrock

Amazon Bedrock is a fully managed service that offers foundational models that can be used and customized to easily build and deploy generative AI applications. The foundation models are offered via a single API, and since it is a managed service, there is no need to manage any infrastructure on your own.

Along with announcing expanded choices for models, AWS also announced the possibility to put guardrails around your models. The guardrails assess user inputs and model responses based on the policies that you provide to ensure that your AI-powered apps adhere to data privacy and responsible AI standards based on your specific use case. For example, you can use the guardrails to safeguard the context of your application, restrict undesirable topics or filter harmful content for a relevant and safe user experience.

Additionally, model evaluation for Bedrock is now available in preview. It offers a choice of human evaluation or automatic evaluation based on predefined or custom metrics to evaluate, compare and choose the best models for your use cases.

2. New AWS Trainium Chip for AI Models

During the conference, a second generation of the custom-built Trainium chip was announced. Trainium is a family of processors built by AWS for training models, designed to enhance performance and efficiency. The new Trianinum2 has three times the memory as the first chip, and can train models four times faster than its predecessor. 

Together with the other announcements, AWS demonstrates a deep understanding of the advancements and needs in the field of GenAI and ML, and they showcase a commitment to provide the best tools possible for the users to accelerate the training of models as well as building and managing their AI applications.