To deliver the message to the markets, Garman called on a surprise appearance by Amazon CEO Andy Jassy. On Tuesday, December 3 (local time), he announced a new generation of foundation models – Nova.
He announced six variants of the Nova family of models:
Nova Micro: Text-to-text model, with low latency
Nova Lite: Multi-modal model for processing image, video, text inputs.
Nova Pro: Multi-modal mode, USP – accuracy and speed.
Nova Premier: Multi-modal model, USP – most capable of complex reasoning tasks.
Other than these four multi-modal models, Jassy also announced:
Nova Canvas: Which will help generate images
Nova Reels: Video generation model, with studio quality (6 secs up to 2 mins)
At the very outset, Jassy claimed that the Nova family of models outperform contemporary offerings in Google’s Gemini family, Microsoft’s ChatGPT family and Meta’s Llama.
The Amazon CEO also claimed that the image and video outputs generated will be watermarked to ensure traceability and to help identify synthetically generated content.
Giving a glimpse into the future, Andy also revealed that Amazon was working towards AI models that will be able to process prompts across formats (text, images, audio, video) and generate responses across formats (text, images, audio, video).
These foundation models will be integrated with AWS offering, Amazon Bedrock. AWS customers can access leading Gen AI models through Bedrock, with a single API.
