Oracle CloudWorld 2023: Ellison Calls AI ‘A Revolution’ In Keynote

‘This makes AI central to almost everything we’re doing,’ Oracle co-founder Larry Ellison says.

Oracle co-founder and Chief Technology Officer Larry Ellison said that generative artificial intelligence has been an opportunity for his company, highlighting a supercomputer his company is building with Nvidia and Oracle AI use cases in health care and software development.

“Generative AI is a revolution,” Ellison said during his keynote address at Oracle’s annual CloudWorld event. “It is a breakthrough. It’s transformational. It’s fundamentally changing things at Oracle. … This makes AI central to almost everything we’re doing.”

Customers are clamoring for GenAI because of the security benefits from removing humans from processes, Ellison said. But he predicts that not many customers will want foundational, large language models like those made by ChatGPT creator OpenAI, backed by Oracle rival Microsoft.

Oracle’s Ellison Talks GenAI

“They want to take these standard foundational AI models then you want to specialize them for law,” Ellison said. “Or for medicine. Or for some other field. They want to use their own training data to specialize this. And the way you do that, the best way to do that, is to put that supplemental training data in an Oracle vector database. And that’s what we built. Because we can move that data into the training machines faster than anybody.”

CloudWorld 2023 runs through Thursday in Las Vegas.

Ellison called the supercomputer Oracle is building with Nvidia “the largest scientific computer ever built in the history of the earth.” In software development, the ability to produce code with Oracle application Apex means Oracle developers won’t handwrite apps moving forward.

“If we’re starting a brand new project, we’re generating that code. We’re not writing it. We’re not handwriting it anymore,” he said.

And in health care, Ellison said that doctors will turn to GenAI for writing drafts of orders and discharge notes – emphasizing that the AI does not take over the doctor’s job. “The model is making the doctor’s job easier,” he said.

Like other GenAI vendors, Oracle has emphasized data privacy when training AI models.

Salesforce co-founder and CEO and one-time Oracle employee Marc Benioff even railed against other AI vendors during a keynote address this month.

“In the Oracle Cloud, your training data, you can train and specialize a private model with your private data that remains private after training,” Ellison said. “And we’ve done that for software companies, financial services companies, medical companies.”

And Ellison is pushing for more adoption of Oracle Autonomous Database by customers and internally. More automation won’t result in layoffs, however, Ellison said – bringing to mind comments made by IBM CEO Arvind Krishna earlier this year.

“A small percentage of Oracle databases in the world are our new Autonomous Database,” Ellison said. “That’s going to change. It’s going to become a much larger percentage. … Even inside of Oracle. It wasn’t used by Fusion. It wasn’t used by NetSuite. That’s all changing. All of our new applications are starting with the Oracle Autonomous Database.”

Ellison continued: “Does that mean we’re going to have massive layoffs here at Oracle? No, we’re too ambitious for that. What we are is going to try to do a lot more.”

Here’s more of what Oracle’s co-founder had to say during CloudWorld 2023.

Oracle’s AI Opportunity

How is this (the growth of GenAI) working out for Oracle? So far, we’re pretty happy.

It turns out that Oracle’s Generation 2 Cloud is very different from other clouds in a number of ways. But one of the most interesting ways is that the network we use to interconnect our computers in the cloud is very different than what the other cloud vendors provide.

We provide a remote data memory access network – that means one computer in the network can actually access the memory of another computer without tapping that computer on the shoulder and getting it to interrupt itself.

So it has the ability to move a lot of data from one computer to another extremely fast. Many times faster than conventional networks.

And the fact that our standard network and all of our clouds use RDMA network means that when we build a computer for training large language models – that would be a computer made of Nvidia GPUs (graphics processing units).

When we interconnect those GPUs, that computer runs much faster in our cloud than it does in other clouds.

Well, as I’ve said before, in the cloud, time is money. If we run twice as fast, we cost half as much. We run three times faster, we cost a third as much. We are much faster and many times less expensive than the other clouds for training AI models.

That’s why Nvidia is doing AI training in the Oracle Cloud. Cohere is doing AI training in the Oracle Cloud. (Tesla CEO and Twitter owner) Elon Musk’s new xAI is doing training in the Oracle Cloud. Plus dozens of additional technology leaders and startups are coming to the Oracle Cloud because it’s faster.

Much faster and much cheaper. Much more economical to build your AI in the Oracle Cloud.

Oracle Working On Supercomputer

The Nvidia superclusters we’re building, well, one of them will be the largest scientific computer ever built in the history of the earth.

I remember when I used to keep track of where the world’s fastest computer was. Was it in the United States? Was it in Japan? Where was this computer?

Was it built by Cray in the old days or by Intel. Who built it? Well the new answer will be Nvidia and Oracle. And it’s a bunch of Nvidia H100s … connected together with our RDMA network.

Now there are two things that make a computer fast when you’re dealing with AI. One is how fast are GPUs – how fast you process the data.

The other is how fast can you move the data to the computer to process it. And we do that much, much faster.

And if you’re not moving the data fast enough, the processors are just waiting. And our big advantage is we move data better than anybody else. Because we started using that RDMA network for our database, which specialized in moving data around.

So happens that’s the same problem you have with GPU clusters, moving data around, it’s a problem we solved for databases and we use in building AI models.

Oracle GenAI For Coding

Generative AI is a revolution. It is a breakthrough. It’s transformational. It’s fundamentally changing things at Oracle. … This makes AI central to almost everything we’re doing.

And it fundamentally changes how we build applications. How we run applications. … For example, we’re not going to be writing new applications anymore in Java.

Not new ones. I mean, we’re continuing projects that started in Java awhile ago. And they will go on for a very long time. We’re not converting from Java to anything else. We’re continuing to use Java.

But if we’re starting a brand new project, we’re generating that code. We’re not writing it. We’re not handwriting it anymore.

We’re generating that code in this thing called Apex. And this application generator’s a no-code system. Because of new technology we have improved Apex dramatically over the last couple of years to the point now where virtually every one of our new applications will be generated by Apex. … A small percentage of Oracle databases in the world are our new Autonomous Database.

That’s going to change. It’s going to become a much larger percentage. … There’s been a long debate about the Oracle database and relational database. Is that the best way to go? Or really should (you) use an object database like Mongo? And use conventional … inputs and not worry about this schema that thing you have to build.

Well, we’ve solved that problem. We’ve ended that debate. … We’ve added vector capability to our standard database because most people want to take these AI models – what are called foundational models, the ones built by Cohere and OpenAI, xAI and others – they want to take these standard foundational AI models then you want to specialize them for law. Or for medicine. Or for some other field.

They want to use their own training data to specialize this. And the way you do that, the best way to do that, is to put that supplemental training data in an Oracle vector database. And that’s what we built. Because we can move that data into the training machines faster than anybody.

Oracle Apex Improves Security

What used to be called our analytics platform is now a data intelligence platform because it is looking at the data using AI and then it itself is the repository for so much data.

You’re using that data intelligence platform as a source of data to specialize large language models. … Apex, Java programming to re-generating new applications. Cerner’s new Millennium (electronic health record platform), generated with Apex.

Fusion Marketing, generated with Apex. … Banking, retail, hospitality. Virtually every new project is being built to be generated with Apex. That means the teams are dramatically smaller. The development process is fundamentally different. We prototype features and iterate so we develop things much more quickly.

When we generate an application, guess what? We don’t generate security bugs. Every one of our apps that we write has to go through a security audit. Well, not the Apex app. Because we didn’t write it. The AI system wrote it. We generated the app. It doesn’t generate security bugs.

And it generates an application that is … stateless. … If the computer it is running on goes down, you can immediately failover to another computer. So it’s fault tolerant. They’re much more reliable.

So as long as you’re willing to spend fewer people and less time building your application, it will be more secure and more reliable and have a better UI (user interface) and all of those things. This is a very big deal. We’ve been programming in Java for a long time.

More Oracle Autonomous Database Adoption

The Autonomous Database … it was not used by a majority of our customers.

Even inside of Oracle. It wasn’t used by Fusion. It wasn’t used by NetSuite. That’s all changing. All of our new applications are starting with the Oracle Autonomous Database. And Fusion and NetSuite are converting to the Autonomous Database.

Everything we have that’s current, everything that we built a while ago, everything that’s strategic and on the cloud will be Autonomous Database because it’s so much more secure.

If you don’t have human labor, yeah, you save money. But all the security errors – almost all the security problems – are caused by human error. And if there is no human labor, there is no human error.

The only way you can build a secure system and truly secure system is to eliminate human error. Self-driving cars … they’ll crash 1 percent as often, a tenth of a percent as often as human-driven cars.

Automation is much more reliable, much more predictable, much safer than doing it by hand.

More Customers Will Seek Specialized Models

Most customers are not going to be building the same thing that xAI is building or OpenAI builds or Cohere builds. They’re not going to be building one of these foundational, large language models.

They’re not going to be trying to solve the self-driving problem and process vast amounts of image data and build a special neural network just to handle the image data.

They’re not going to be doing that. … You can come to the Oracle Cloud and start, let’s say, with a Cohere foundational model and then take supplementary data.

Let’s say we take the Cerner EHR data – anonymized, of course, anonymize it – and we use that EHR, that medical data, to specialize Cohere AI model on medicine. So we train it on medicine. And that’s exactly what happened. A company in Israel called Imugene actually did that. They took a bunch of cancer biopsy slides and fed that into the AI model, and the computer is now able to diagnose cancer in a matter of minutes. … A lot of people will be building these specialized models.

Not many companies are going to build foundational models. … We are taking a lot of electronic health data and training it to give doctors orders, whether it’s discharge notes … the idea is to give the doctor a draft of a discharge note or a draft of an order.

The doctor reviews it, edits it and then submits it. So the model is making the doctor’s job easier. It is not taking over their doctor’s job.

Oracle Approach To Training Data

A lot of people are very concerned about sharing their training data. Let’s say we have an investment banker that’s a customer. And they have a bunch of financial trading data that they don’t want to share with the world. But they’d like to use it to train a model.

Well, in the Oracle Cloud, your training data, you can train and specialize a private model with your private data that remains private after training.

And we’ve done that for software companies, financial services companies, medical companies. … I’m not sure if there’s anything more important that we’re working on right now then our new Oracle Cloud Data Intelligence platform.

That’s a combination of Oracle Analytics plus generative AI. And the specific example I’d like to cite is Oracle’s new public health data intelligence platform. … This started as a project called Cerner Healthy Intent. But we’ve rewritten, rebuilt it since we bought Cerner. And it’s become the Oracle public health data intelligence platform.

And we’ve unified national population scale health data. This is designed to take all of the EHR data for a country, for all the patients in a country and put it together.

All the diagnostic laboratory data. Everything. Putting that into a single Oracle Autonomous Database for the entire population of the country.

When you do that, when you take all of this health data and put it in one place, you get enormous benefits. The first benefit is when you go to train AI models, you have 1,000 times more data than you used to have. You have all the data. … This is like a clinical trial that goes on forever. The clinical trial for everything that goes on forever.

We keep collecting all of this data. And then we use that data to train models. … You’ll be shocked that some data still isn’t digitized. … Whether they’re MRIs … Siemens scanners, even sonograms, sonograms aren’t saved. Biopsies sit on glass slides in drawers. They’re not digitized by and large. … We don’t capture this data and digitize it and vectorize it for training data. And we should. And we’re in the process of doing that.

Automation Doesn’t Mean Layoffs

What does it mean when I say we’re going to generate applications rather than write them in Java. Teams are going to be smaller and they’re going to write applications faster.

Does that mean we’re going to have massive layoffs here at Oracle? No, we’re too ambitious for that. What we are is going to try to do a lot more.

Because we can generate programs, we can tackle things that are much bigger in scale than we’ve been able to tackle in the past.

We have better tools for building applications, we can build better applications faster. … The latest generation of greenhouse is being developed in Southern California. … Applied Invention is using the Oracle Cloud, the Autonomous Database, NetSuite to manage the business and collaborating with the Oracle engineering team on AI, IoT (internet of things), robotics, analytics.

Oracle Builds A Police Car

Our next generation police car is coming out very soon. It’s my favorite police car. It’s my favorite car actually. It’s Elon (Musk)’s favorite car. It’s incredible.

I know too much about it. Some of it is still to be disclosed. But among other things, it’s very safe. Very fast. It’s got a stainless steel body. And we don’t have to add … cameras to it because already we actually use their existing cameras and their existing screen to put our application up for people in the Tesla vehicles. … It’s actually deployed in Stanislaus County, I know, for the police. And I believe, also, for the fire (department). … Stanislaus County is a rural area in California near Yosemite Valley.

And of course, it’s an area … vulnerable to brush fires. … Firefighters are in remote areas where there is no cell service. So we have to communicate with firefighters by using satellites. And we have to inform firefighters in a safe way using, again, robotic drones. And the drones are fabulous during the dry season.

They can be used for monitoring firefighting and fire monitoring for forest fires. They can be used in urban areas during heavy traffic. They provide improved situational awareness. Very important during fires to know what streets are open and what streets are closed and know immediately and be able to let people know immediately what’s the best way to exit the area.

Call For End Of Data Egress Fees

Last week I was in Redmond, Wash., with Satya Nadella, the CEO of Microsoft. And we had a wonderful chat.

And we agreed on one big idea. And that big idea was – clouds should be open. They should not be walled gardens. Clouds should be interconnected.

And customers already use multiple clouds. People don’t go to one and only one cloud. In infrastructure, there are four hyperscalers. AWS, Microsoft Azure, Google and us.

And there are a lot of application clouds. … It’s our job, once customers choose what they want to use, to make all of that stuff work seamlessly together. To interconnect those clouds in such a way that if you want to use the Salesforce sales automation application and then have a data warehouse in Oracle, for example, where you mash up that data with other data, that that’s a very simple, straightforward thing to do. Because the two clouds are connected.

If you want to use OpenAI’s Chat GPT with an Oracle Autonomous Database … any of the Microsoft services with Chat GPT, with Oracle, that should be easy to do. There shouldn’t be a wall between Microsoft Cloud and the Oracle Cloud.

There shouldn’t be a charge if you want to move your data out of the AWS cloud and put it into a database in the Oracle Cloud. That is your data.

Call For Open Systems

And so those clouds should be interconnected. And they should seamlessly interoperate. And we used to call that open systems. And we think – Satya and I think – the world’s going back in that direction. Customers are insisting on it because they use multiple clouds.

We (Microsoft and Oracle) started experimenting with this back in 2019. We internet-connected dark clouds. We had a lot of successful customers. … But it wasn’t that easy to use. … You had to be fairly sophisticated to connect Microsoft services to Oracle’s services.

We decided to make it completely transparent. So you can go straight to the Azure portal and configure an Oracle service like the Autonomous Database and connect it to a Microsoft service. And make it really fast. So (what) we decided to do was actually build Oracle clouds right inside of Azure data centers. … And there are no charges. No data egress fees. It’s your data. It is your data, whether it’s in an Oracle Cloud, in a Microsoft Cloud, in an AWS cloud, in a Salesforce cloud. It is your data. They shouldn’t charge you for moving it. … We’re building, I think, a dozen of them. A dozen data centers in Azure data centers with Microsoft.

We’re building them at specific customer sites. A couple of big phone companies are getting them to run their network. … And right now, we think we have more data centers than any other hyperscaler.

But still, we’re on our way to 100. We think, looking at the price of the DRCC (Dedicated Regon Cloud@Customer), we’re going to be the hyperscaler that measures its data centers not on the way to 100, but really on the way to 1,000.

And this thing connects to all the other clouds. So it will connect to the Salesforce Cloud. Connect to the Microsoft Cloud. It will be your interconnect to all of the other clouds. Plus there’ll be your own dedicated cloud region, fully automated, fully modern, RDMA network, everything – at a very low cost that may fit your budget.