NVIDIA unveils AI Foundations, a cloud service totally focused on artificial intelligence

NVIDIA makes big announcements around a dedicated AI cloud service. AI Foundations will serve numerous purposes.

Everyone is talking about artificial intelligence. It is on everyone’s lips, and the general public is just beginning to taste the power of these technologies. This prompts tech giants to dive into it fully. For example, NVIDIA unveils AI Foundations, a cloud service that will allow businesses to easily and quickly develop their models “to design, refine, and exploit custom language models and generative AI models that are trained with their own data and created for very specific tasks.”

NVIDIA makes big announcements around a dedicated AI cloud service

Among these models are NeMo, NVIDIA’s text-to-image generation engine, a competitor to DALL-E 2, and BioNeMo, a version oriented towards drug and molecule research derived from NeMo. There is also Picasso, an AI for generating images, videos, and “3D applications… to boost productivity for creativity, design, and digital simulation.” NeMo and BioNemo are available in early access, while Picasso is in private beta. All three run on the NVIDIA DGX Cloud platform and will have a dedicated online portal. These models can be trained with 8 to 530 billion parameters, more than triple the 185 billion parameters of GPT-3.5.

Imagine StableDiffusion, but trained on Getty Images with Getty’s permission. NVIDIA announced such a system built on the NeMo cloud service: a series of carefully selected text-to-image and text-to-video models, “trained on content from Getty Images.” “Getty Images will pay royalties to artists on all revenues generated by these models.” BioNeMo uses the same techniques as NeMo but for molecule and drug research, “allowing researchers to refine their generative AI applications on their own proprietary data and run AI models directly in a browser or via the API of this new service for easy integration into existing applications.”

AI Foundations will serve numerous purposes

Six models will be available at launch, including DeepMind AlphaFold2, Meta AI ESM2, and the predictive models ESMFold ProGPT-2, DiffDock, and MoFlow. According to these companies, incorporating AI-based predictive models helps reduce training time for molecule analysis and optimization from three months to just four weeks.

NVIDIA also announced a similar partnership with Shutterstock. The site will use Picasso to generate 3D objects via simple textual inputs as part of a new Creative Flow feature. This should be available on Turbosquid.com and its future Omniverse platform.

NVIDIA is also collaborating with Adobe for its Content Authenticity Initiative, aimed at improving transparency and accountability in the generative AI training process. Among the proposals are the creation of a “do not train” list, like a robot.txt but for image and multimodal content, and tags that would explicitly indicate if content has been generated by an AI and which one. The two companies also announced that many of Picasso’s features will be integrated directly into Adobe’s software suite, including Photoshop, Premiere Pro, and After Effects.