News

New Nova Foundational AI Models Lead AWS re:Invent 2024 News

• By David Ramel
• 12/03/2024

This week's AWS re:Invent event saw a plethora of AI-related news from cloud giant Amazon for its cloud computing platform, including new foundational AI models.

Those would be the Amazon Nova family, reportedly coming with state-of-the-art intelligence across a wide range of tasks along with industry-leading price performance.

The new family of models follows the trend of moving from huge all-in-one, jack-of-all-trades constructs to more specialized models designed to more nimbly handle specific tasks.

Thus four new models introduced to work with the Amazon Bedrock AI service include:

• Micro: Text-only, fastest and cheapest, ideal for tasks like summarization and basic reasoning.
• Lite: Low-cost multimodal model, handles real-time customer interactions, document analysis, and visual question answering.
• Pro: Most capable multimodal model, excels in complex tasks like financial document analysis and video analysis. Integrates with APIs and tools for complex workflows.
• Premier: Most powerful for complex reasoning tasks, also serves as a teacher model for distilling custom model (coming in Q1 2025, while the others are available now).

The company also announced two creative content generation models:

• Amazon Nova Canvas: This generates high-quality images from text descriptions. It offers precise control over style, content, and editing features like inpainting, outpainting, and background removal.
• Amazon Nova Reel: This creates professional-quality short videos from text prompts and images. It allows users to control visual style, pacing, and camera movement.

As mentioned, the models work with Bedrock, a fully managed service from AWS designed to help developers build and scale generative AI applications.

"Integration with Amazon Bedrock makes deployment and scaling straightforward," the company said in a Dec. 3 blog post. "You can leverage features like Amazon Bedrock Knowledge Bases to enhance your model with proprietary information, use Amazon Bedrock Agents to automate complex workflows, and implement Amazon Bedrock Guardrails to promote responsible AI use. The platform supports real-time streaming for interactive applications, batch processing for high-volume workloads, and detailed monitoring to help you optimize performance."

Looking forward, speech-to-speech and multimodal-to-multimodal functionality is in the works:

• Speech-to-Speech Model: Coming in early 2025, this model will revolutionize conversational AI. It will understand and respond to natural language speech, considering factors like tone and cadence, to provide more human-like interactions.
• Multimodal-to-Multimodal Model: Set to launch in mid-2025, this model will be capable of processing and generating content across various modalities, including text, images, audio, and video. This will enable the development of AI agents that can understand and respond in a wide range of ways, simplifying complex tasks and opening up new possibilities.

The company also announced AI news around its Amazon Q Developer generative AI-powered assistant for software development and Amazon SageMaker, a fully managed service from AWS that simplifies the process of building, training, and deploying machine learning (ML) models. And all of that is but a small part of the AI-related news, services and products being announced at re:Invent, so stay tuned.