AWS Nova AI models debut with enhanced customer control service
▼ Summary
– AWS announced Nova 2, a new fleet of four upgraded AI models, during its Re:Invent keynote.
– The new models include cost-effective reasoning, complex task handling, conversational AI, and multimodal capabilities.
– AWS also launched Nova Forge, a service allowing enterprises to build custom versions of Nova models for $100,000 annually.
– Nova Forge is designed to help companies train models on proprietary data without losing core reasoning abilities.
– Early customers for the Nova Forge service include companies like Reddit, Sony, and Booking.com.
Amazon Web Services has significantly expanded its portfolio of proprietary artificial intelligence models with the introduction of the Nova 2 family. Unveiled during the AWS Re:Invent keynote, this launch represents a strategic push to provide enterprise clients with more powerful and specialized AI tools. The update includes four new models and a novel service designed to give businesses unprecedented control over model customization, directly addressing a common challenge in corporate AI adoption.
The new lineup features Nova 2 Lite, a cost-efficient reasoning model adept at handling everyday tasks by processing text, images, and video. For more demanding applications, Nova 2 Pro serves as a reasoning agent built for highly complex duties like coding, capable of interpreting text, images, videos, and speech. AWS also introduced Nova 2 Sonic, a dedicated speech-to-speech model for conversational AI, and Nova 2 Omni, a multimodal system that can accept various inputs and generate both text and images.
A major component of the announcement is Nova Forge, a new service allowing AWS cloud customers to develop their own customized versions of the Nova models, referred to as Novellas. This offering, reported to carry an annual cost of $100,000, provides access to models at different training stages, pre-trained, mid-trained, or post-trained. Enterprises can then fine-tune these models using their own proprietary data, a process aimed at preserving core model capabilities while integrating specific business knowledge.
AWS CEO Matt Garman highlighted the rapid adoption of the original Nova models, noting they are now used by tens of thousands of customers ranging from large corporations to innovative startups. He explained that Nova Forge tackles a critical technical hurdle known as “catastrophic forgetting,” where models lose previously learned information when new data is introduced. Garman compared the challenge to an adult learning a new language, which is far more difficult than acquiring language skills in childhood, suggesting that Nova Forge offers a more effective path for late-stage model training.
Early adopters of the Nova Forge service include notable companies like Reddit, Sony, and Booking.com, indicating strong initial enterprise interest. This expansion underscores AWS’s commitment to not only providing foundational AI infrastructure but also delivering tailored solutions that enable businesses to build and control AI systems aligned with their unique operational needs and data environments.
(Source: TechCrunch)
