Amazon has introduced a sequence of recent AI basis fashions beneath a brand new “Nova” branding that can be out there as a part of the Amazon Bedrock mannequin library in AWS.
The corporate can also be coaching a mannequin known as Amazon Nova Premier, which it says can be “our most succesful multimodal mannequin for advanced reasoning duties.” Amazon goals to make Nova Premier out there in “early 2025.”
Amazon is releasing content material era fashions, too: Amazon Nova Canvas, a picture era mannequin, and Amazon Nova Reel, a video era mannequin. The corporate says that these fashions have “watermarking capabilities” to “promote accountable AI use.” For instance of what’s doable with Nova Reel, Amazon has shared this mock advert for a faux pasta model.
Later in 2025, Amazon plans to launch a speech-to-speech mannequin and “a local multimodal-to-multimodal” mannequin, in accordance with a weblog put up.
Amazon introduced these new fashions at its AWS re:Invent convention, which is occurring now in Las Vegas. On the present, the corporate additionally stated that it’s constructing an enormous AI compute cluster that depends on its Trainium 2 chips in partnership with Anthropic (which it has invested $8 billion in). “When accomplished, it’s anticipated to be the world’s largest AI compute cluster reported so far out there for Anthropic to construct and deploy its future fashions on,” in accordance with Amazon.
The corporate, like many different large tech gamers, is racing to launch new AI merchandise and options to remain forward of newer firms like OpenAI. The place Amazon might have a bonus is how a lot web infrastructure is already powered by AWS — giant enterprises could also be extra keen to make use of Amazon’s AI choices as a result of the corporate has already a trusted status. An Apple exec even appeared onstage at present at re:Invent to speak about how the corporate depends on Amazon’s customized AI chips.
