Amazon Web Services (AWS), at its flagship tech event re:Invent, announced a slew of AI innovations. The company also introduced a new AI model capable of processing text, speech, images, and video. The model named Nova 2 Omni is currently available in preview and can also be used to stitch together multiple models. The Nova 2 Omni is part of the Nova family of models first unveiled at last year’s re:Invent and which now also includes Nova 2 Lite and Nova 2 Sonic.
Along with this, the tech giant also unveiled Nova Forge, which allows customers to develop their own AI models. During the keynote speech, AWS CEO Matt Garman said that this approach would lead to better accuracy and a higher degree of customisation.
AWS also announced new innovations in Amazon Bedrock AgentCore, which it said is the most advanced platform for securely building and deploying agents at scale. Though AI agents can reason and act autonomously, there is a need for robust controls to prevent unauthorised actions, and Amazon Bedrock AgentCore with its new capabilities addresses this, as per AWS.
Amazon Bedrock AgentCore with new capabilities
The company said that Amazon Bedrock AgentCore is launching Policy in preview, allowing teams to set clear boundaries for agent actions with natural language. Meanwhile, AgentCore Evaluations simplifies monitoring quality with 13 pre-built evaluators for dimensions like correctness and safety, continuously assessing live interactions to send alerts when performance drops. Another feature, AgentCore Memory, introduces episodic functionality that allows agents to learn from experiences, improving decision-making.
Besides, the tech giant also announced Amazon Bedrock’s largest expansion ever of new models yet. The company has added about 18 fully managed, open-weight new models. These include four Mistral AI models, available first on Amazon Bedrock, from Mistral AI. With the latest development, Amazon Bedrock now offers about 100 serverless models, offering a broad range of models from leading AI companies.
The company also introduced EC2 Trn3 UltraServers, powered by the advanced Trainium3 chip. These Ultraservers offer 4.4x compute performance, 4x greater energy efficiency, and about 4x more memory bandwidth than their predecessors. AWS claims that these technologies have been designed for next-generation AI workloads, as the Trn3 UltraServers allow faster model training, reduced latency, and cost-effective inference at scale. Also, AWS is previewing Trainium4, which promises at least 6x processing performance and 3x FP8 performance, paving the way for even greater advancements in AI innovation.
Frontier Agents and AI factories
The tech giant introduced a new class of AI agents that is a step forward in agentic capabilities. According to the company, these AI agents are autonomous and scalable and can work for hours or days without any intervention. AWS also announced three frontier agents – Kiro autonomous agent, AWS Security Agent, and AWS DevOps Agent. While the Kiro autonomous agent works as a virtual developer, the AWS Security Agent acts as a security consultant. Meanwhile, the AWS DevOps Agent acts more like an on-call operational team.
Story continues below this ad
The company announced AWS AI Factories, its new service that offers enterprises and governments dedicated AWS AI infrastructure deployed in their own data centres. The AWS AI Factories combine Nvidia’s AI computing and Trainium chips, AWS high-speed, low-latency networking, high-performance storage and databases, security, and energy-efficient infrastructure along with AI services such as Amazon Bedrock and SageMaker AI, enabling customers to rapidly develop and deploy AI applications at scale.
© IE Online Media Services Pvt Ltd