Not novel, but Nova
Amazon announces Nova, a new family of multimodal AI. It seems as though there are 6 models included in there recent creation. One of the models, Nova Canvas, is specifically for image generation, and another Nova Reels, is dedicated to video. The other four models are text generators (Micro, Lite, Pro and Premier).
The text generating models speak 15 languages, but excel at English. Despite being the fastest, the Micro model can only input and output text values. The Lite, Pro and Premier models can process images, video and text input. In addition to the multimedia capabilities, Premier can assist with creating tuned custom models.
Amazon claims that the models are accurate and the fastest in their class, while also being the least expensive to run. They can be accessed through Amazon’s AI development platform, Bedrock, where they can also be fine tuned to meet client needs. The three multimedia models also have agent capabilities, allowing them to take a few independent steps on their own, after being trained how.
There are safety measures baked into the models to embed watermarks, for copyright purposes, as well as moderation tools to combat the spread of misinformation, child sexual abuse material, and risks involved with weapons that lead to war crimes.
It’s apparently a trade secret to access the parameters, weights and training data, used for the Nova models, but AWS claims that it’s a combination of proprietary and licensed data. It’s pretty common to keep such cards close to the chest, as doing otherwise would open the door to increased competition and possibly even litigation over Intellectual Property. Instead Amazon will cover customers responsibility if the model ends them up in court.
Further down the development pipeline we can expect a speech to speech model, that is sensitive to tone, inflection and cadence, as well as a model that will accept and output any media format. It’s a strong addition to a solid foundation at Amazon, and we can’t wait to run it through it’s paces. Until we get a chance to drive it around the block, here are a few of the outputs from Nova so far.
https://techcrunch.com/2024/12/03/amazon-announces-nova-a-new-family-of-multimodal-ai-models/
Does Anthropic know? Claude might be hurt angry or distraught. Can you imagine having to parse a machine’s feelings into an equation so that their pain and anguish don’t muddle or ruin the results?