Nvidia CEO Jensen Huang: AI language models as-a-service “potentially one of the largest software opportunities ever”


jensen-huang-gtc-2022-crop-for-twitter

Nvidia co-founder and CEO Jensen Huang opened the company’s fall GTC conference by announcing the general availability of the company’s new “Hopper” GPU next month in systems from Dell and others. The keynote also presented computers for healthcare, robotics, industrial automation and automotive applications, as well as several cloud services, including a cloud service hosted by Nvidia for deep learning language models such as GPT-3.

NVIDIA

As reported yesterday, Nvidia co-founder and CEO Jensen Huang kicked off his company’s fall GTC conference with numerous product and service announcements, including the launch of two cloud computing services that the company will operate.

In a press briefing on Wednesday, Huang told ZDNET that the two services will be “very long-term SaaS platforms for our company.”

One service, Large Language Model Cloud Services, lets a developer take a deep-learning artificial intelligence program like GPT-3 or Nvidia’s Megatron-Turing 530B and tailor it to specific applications to make it specific to a task while still providing the Reduce effort a customer needs to do.

The second service, Omniverse Cloud Services, is an Infrastructure-as-a-Service offering from Nvidia that allows multiple parties to collaborate on 3D models and behaviors.

Also: Nvidia CEO Jensen Huang announces availability of “Hopper” GPU, cloud service for large AI language models

ZDNET asked Huang: How big can SaaS become [software-as-a-service] Business for Nvidia for many years?

Huang said it’s difficult to know, but that the large language model service has such broad applicability that it will be one of the biggest opportunities in all of the software.

Here is Huang’s answer in its entirety:

Well, it’s hard to say. That’s really, sort of, the answer. It depends on what software we offer as a service. Maybe another way of taking it is just a pair at a time. In these terms and conditions we have announced new chips, new SDKs and new cloud services. And that’s what you’re asking about. I have highlighted two of them [cloud services]. One of them is large language models. And if you haven’t had a chance to delve into the effectiveness of large language models and their impact on AI, please really do. It is really important. Large language models are difficult to train, and applications for large language models are diverse. It has been trained on a great deal of human knowledge. And so it has the ability to recognize patterns, but it also contains an encoded amount, a large amount of encoded human knowledge, so if you will, it has a kind of human memory, if you will. In a way, much of our knowledge and skills are encoded in it. So if you wanted to adapt it to do something it was never trained to do — for example, it was never trained to answer questions or it was never trained to summarize a story or break the news, to paraphrase it, it was never trained to do it those things – with a little extra practice, you can learn those skills. This basic idea of ​​fine-tuning, adapting to new skills, or zero-shot or little-shot learning has major implications for a variety of fields, which is why so much funding is being devoted to digital biology. Because large language models have learned to structure the language of proteins and the language of chemistry. And so we set up this model. And how big can this chance be? My feeling is that every single company in every single country that speaks every single language probably has dozens of different skills that their company could adapt to make our grand language model work. I’m not entirely sure how big this opportunity is, but it may be one of the greatest software opportunities of all time. And that’s because automating intelligence is one of the greatest opportunities of all.

The other opportunity we talked about was Omniverse Cloud. And remember what omniverse is. Omniverse has several properties. The first feature is that it can take, store and assemble physical information, 3-D information across multiple layers or so-called schemas. And it could describe geometries and textures and materials, properties like mass and weight and so on, connectivity. Who is the supplier? How much does it cost? What is it related to? What is the supply chain? I’d be surprised if – behaviors, kinematic behaviors. It could be behaviors of artificial intelligence. So the first thing Omniverse does is store data. Second, it connects multiple agents. And the agents can be humans, can be robots, can be autonomous systems. And the third thing it does is that it gives you a glimpse of this new world, or in other words, simulation engine. So basically, Omniverse consists of three things. It’s a new kind of storage platform, it’s a new kind of connection platform. and it’s a new kind of computing platform. You could write an application on Omniverse. You can connect other applications through Omniverse. For example, we have shown many examples of how Adobe is connected to Autodesk applications connected to different applications. So we connect things and you could connect people. You could connect worlds, you could connect robots, you could connect agents. That’s the best way to think about what we’ve done with Nucleus [Nucleus Cloud, a component of Omniverse Cloud, is a facility for developers to work on 3-D models using the Universal Scene Description specification], consider it the easiest way to monetize that is probably like a database. So it’s a modern database in the cloud. Aside from being in 3-D, this database connects multiple people.

So those were two SaaS applications that we set up. One is called the large language model. The other is basically Omniverse, or a database engine if you will, which we’re going to put in the cloud. So, I think those two announcements — I’m really glad you asked — I’m going to get ample opportunities to talk about them over and over, I’m going to talk about them over and over, but these two SaaS platforms are going to be very be long term SaaS platforms for our business and we will run them on multiple clouds and so on and so forth.



Source link

Also Read :  Vecow launches NVIDIA Jetson AGX Orin-based EAC-5000 Series Edge AI computing system