please welcome to the stage Larry Ellison. Hi, everybody. Okay, let's see. I just took some cold medicine about an hour ago. So if I do sit down on stage, it's not because I tore my Achilles. Okay. I'm gonna miss Aaron Rodgers this year. Anyway, I'm going to talk about. I'm not going to talk about the NFL. Gener of AI showed up about a year ago now, and it changes everything. It certainly changing everything at Oracle. Okay, about a year ago, open AI demonstrated chat GPT, and much to the shock of the people who developed chat GPT, the baby talks. Hope seriously, the developers did not expect that. They did not expect that, and I'll explain why. Opening eye was a company founded about 10 years ago by Elon Musk, and what they did after, again, about a decade's worth of work, is they built this enormous artificial neural network made up of billions of artificial neurons or what AI professionals call a billion parameter model, and no one ever built a model.

quite this large before and after they built the neural netor Network. They trained it on enormous amounts of data. It read all of the Wikipedia. It read the entire public internet. No one had ever trained a model on that much data before. But the artificial neural network itself was not different from the one the previous one or the one before that. It was the same architecture. The only difference with GPT 3.5 is the neural network was bigger. The training data was more vast. So just because of this change in scale, suddenly when prompted with questions, chat GBT supplied answers. It engaged in conversations. It was a big surprise for the people at JGPT, for AI professionals, for the entire world. And it captured our imagination. Unlike most events in technology, most cool new tech does not get the attention of heads of state and heads of government and everyday people in other professions. Just doesn't. This is like Sputnik. Now that shows I'm a little bit older than some of you in the audience.

uh Sputnik. When the Russians first La launched the first arti official satellite around the world, it captured our imagination. It was, it was news everybody knew about it. Everybody knows about generative AI. It raised the concerns of AI professionals who were concerned about the risks this new technology poses. It raised concerns about governments who've talked about regulating it. The writer strike not far from here in Hollywood is partially based on the writers being concerned about their jobs. Who's going to write the next great script in Hollywood? Is it going to be some guy, you know, or some computer you don't? Everyone wants to know what comes next. Well, I'm going to start by telling you what I think comes next, which is a worldwide race to build what comes next, to build better AI, still better AI, which will deliver a better future and will have significant Associated risks like all new technologies, from the First Technology fire, which can be misused and Spears, which can be misused to later techn technology like nuclear power, which can can be misused.

all new technologies can be misused. However, by and large, technology and advances in technology have made our lives better, has made us as human beings more prosperous and more comfortable. Generative AI is the most important new computer technology ever. Probably one thing's for certain: we're about to find out because countless billions of dollars are being invested in generative AI and large language models. Chat GPT 3.5 is no longer the current best thing out of Op AI. Chat GPT 4.0 just this past in the past 12 months. Cohere, Chat GPT, other companies have developed new large language models which generate not only language, but they generate images, they write music, they write poetry, they generate computer code. That's getting a little close to home and coming soon and very soon in the next 12 months: complete self-driving cars from Tesla. There's a new drug that's entering phase one clinical trials. The drug was developed in China. It's an antiviral drug. It's not a vaccine. It's

therapeutic that treats COVID-19, all variants of COVID-19, every single one. But it doesn't stop there. It also uh cures cousins of COVID, SARS, and MS. It's a small molecule designed by a computer that binds to a protein on that virus and prevents it from replicating. I told you AI is going to make our lives better in a lot of ways. There are associated risks. At Oracle, we're working on a voice digital assistance for, you know, for clinical purposes. So, I'm once, you know, walking into Stanford and, uh, to go look at my x-rays and they couldn't find them and they brought in one, two, three specialist IT specialist to help, help the, uh, the doc find my x-rays. Now you just say, Find Larry Ellison's latest x-rays. Oh, by the way, show me where the bone is broken. There are more big surprises coming, lots of them, and most of the news is going to be good. All right, so how is this working out for Oracle so far? We're pretty happy.

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 the network we use to interconnect our computers in the cloud is very different than what the other Cloud vendors provide. We provide an a remote data memory Access Network. That means one computer in the network can actually access the memory of another computer without kind of tapping that computer on the shoulder and 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 in all of our clouds uses our RDMA network uh means that when we build a computer for training large language models, that would be an a computer made of Nvidia gpus. 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, it costs

half as much if 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. Coh here is doing AI training in the Oracle Cloud. 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 OR Cloud. So, so far so good for Oracle. The Nvidia superclusters we're building, well, one of them will be the largest scientific computer ever built in the history of Earth. I remember when I used to keep track of where the F world's fastest computer was. Was it in the United States? Was it in Japan? Where, you know, where was this computer? Was it built by Cray in the old days or by Intel or who, who built it? Well, the answer, the new answer will be Nvidia and Oracle and

it's a bunch of Nvidia h100. 16,000 of them 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 do 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 and 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 database. and reused in building AI models. okay. Oracle's been using AI for a very very long time. many years. but this is different. generative AI is a revolution. it is a breakthrough. it's transformational. it's fundamentally changing things. at Oracle. I'm GNA see if this works. I'm going to

bring a chair over. Forgive me, this is great. Okay, I'm happy. I like this. Now I don't have to get down on the ground like on Rogers. All right, so I was thinking about that, you know, my family's watching on the internet. They're going to, you know, like what I'm G to get calls after this. Okay, so we have been using AI for a long time, but this makes AI central to almost everything we're doing, and it fundamentally changes how we build applications, how we run applications. It changes everything. 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 a while 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 is a

no Code system and we have imp because of the new tech. 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 is going to be generated by Apex, the autonomous database, which is that a 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 uh the Oracle database. There's been a long debate about the Oracle database, a relational database. Is it is that the best way to go? It's uh or we really should use an object database like and uh use use conventional CH gets inputs and not worry about this scheme of thing you have to build. Well, we've solved that problem. We We've Ended that debate and I'll explain in a minute. Uh We've added Vector capability to our standard database because uh most people want to take these AI models, what are called foundational models, the ones built by coher and open Ai xai and others. They want to take these standard

foundational AI models. Then, they want to specialize them for law, or for medicine, or for some other field, and they want to use their own training data to specialize this. And the way you do that, the best way to do that, is 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. And the next thing is our data intelligence, our analytics platform. 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's 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. I'm going to go, I'm going to deal with each one of these, okay. Apex, from Java programming to generating new applications, CA New Millennium generated with Apex, Fusion marketing generated with Apex, banking applications. Back up the slide, please. I didn't mean to hit the button.

fault my bad but please back it up one okay the soon New Millennium Financial uh Fusion marketing banking retail Hospitality virtually every new project uh is being built or ex me generated with Apex. That means the teams are dramatically smaller, the development process is dramatic is fundamentally different. Uh we prototype features and iterate, uh so we develop things much more quickly when we generate an application. Guess what, it doesn't. We don't generate security bugs. Every one of our apps that we 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, it doesn't. And it generates an application that is quote stateless. Without going into all of that, if the computer it's running on goes down, you can immediately fail over to another computer, so it's fall 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 more secure and more.

reliable and have a better UI, and all of those things. This is a very big deal. We've been programming in Java for a long time. The autonomous database again, it's very cool. I mean, it's 100% self, uh, self-driving database. You don't need a DVA. It installs itself, configures itself, updates itself, patches itself, tunes itself, does all of that stuff on automatically. Still, it was not used by a majority of our customers, uh, even inside of Oracle. It wasn't used by Fusion, 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. It's because it's so back, 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.

if there is no human labor, there is no human error. The only way you can build a secure system, a truly secure system, is to eliminate human error. Self-driving cars will not crash, or they won't crash, they'll crash 1% as often, a tenth of percent as often, than human-driven cars. Automation is much more reliable, much more predictable, much safer than doing it by hand. The autonomous database is also fully elastic in the cloud. So, let's say you're running a basic accounting system, and once a month, or once every two weeks, you run your payroll. And you're a big company, a big retailer, and you have a few hundred employees, and the payroll takes a while to do all the calculations, all the withholding tax, vacation days, all of this other stuff to do all the calculations, and to get that done quickly, you'd like to allocate even more processors. Well, the autonomous database is completely elastic. If you need more processors, you immediately allocate them, use them, run 10 times faster. When you're done, you

give them back to the pool so it's much faster. You get the processors you need when you need them. When you don't need them, you give them back, so you only pay for what you use. This is not how other Cloud databases work. They don't, you say, I want eight processors or 16 or 32 or 64, and you pay for them whether they're working or not. In our case, it gets what you need, you go faster, returns it. If your database is doing nothing, there are no processors allocated, and it costs nothing, so it's much higher performance, much lower in cost, and because of no labor, much more secure. Uh, object data relational database. The big, the big relational database is much more powerful. The only, the issue that programmers have with relational databases is they have to predefine it, what's called the schema. Again, I'm not going to go into a lot of details, but you have to predefine the schema before you start programming. Programmers say, I don't want to do that. I just want to start coding. I'll, I'll.

Define my user interface. Which are JSON documents or what we call JSON objects and in JSON notation. And there are databases that support JSON. There is JSON data. We have one too. We have one too. But what we do now is we generate the schema from the JSON documents. The programmer doesn't have to worry about the schema anymore. We automatically generate this schema. You just want to worry about your user interface and your documents. Programming will generate the schema and you'll end up with a relational database foundation even though you did nothing. The system will generate it automatically and you have the power of SQL for a query language and you have the simplicity of a JSON just dealing with documents just dealing with your user interface program that you write. Easier to write programs but the programs are durable. You have a powerful query language. You get the best of both worlds and they absolutely coexist because we generate the schema. Lots of things are

changing um most customers uh are not going to be building are not going to be building the same thing that XI building or open AI builds or coher builds. They're not going to be building one of these foundational model 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 they're they're not going to be doing that uh again coher does um you know open AI does xai does but most people are going to take what cohere builds and and they're going to just like sending your kid to college and then after college you send them to law school uh so they're going to start with a college-educated kid from coher and then send him to law school or or send them to medical school to train them specifically on medicine and that's what Oracle we we make easy in the Oracle Cloud you can come to the Oracle Cloud start let's say you know with a coher

foundational model, and then take supplementary data. Let's say we take the, the Cerner EHR data, anonymized of course, we anonymize it, and we use that EHR, that medical data to specialize to cohere AI model on medicine. So we train it on medicine, and that's exactly what happened, a, a company in Israel called Imaging, actually did that. They took a bunch of cancer biopsy slides and FedEd in into the AI, into the AI model, and they're now able to, the computer is now able to diagnose cancer in a matter of minutes. I don't know if, imagene is here Cloud world, but it's a remarkable team from Israel that's done absolutely brilliant work. So, it's, this, but this will be a more normal thing to do than a lot of people will be building these specialized models. Not many companies are going to build foundational models, and that makes, that makes complete sense. We're doing the same thing by the way, we're not

taking biopsy slides, and trying to train a model for cancer detection. But we are taking a lot of electronic Health Data and training it to give Doctors Orders, whether it's discharge notes. But again, 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 the doctor's job. The idea is this is a very powerful tool that makes the physician's job easy here, whether the physician is diagnosing something using Imaging or placing an order for a patient in a hospital using C Millennium. Okay, one of the things we do is we take the image, the biopsy data from Imaging. They keep it private. They keep it in our Cloud. But one of the interesting things is we can't see it. No one else can see it.

private to Imaging. So a lot of people are very concerned about sharing their training data. Uh, let's say, 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 like to use it to train a model. Well, in the Oracle Cloud, your training data, you can, you can train and specialize a private model, uh, with your private data that remains private after training. And we, we've done that for software companies and for, uh, financial services companies, medical companies, and it's in fact, I last year, I said, uh, the reason we're optimistic about being able to make real progress in automating, uh, the world of medicine, which is enormous, is that we're not trying to do it alone, and it's important that we have partners like Ronin, uh, working on part of the problem, and partners like an Imaging working on other parts of the problem, and we provide some of the fundamental tools that make their job easier. Another thing I'm not sure

there's anything more important that we're working on right now than uh our new uh Oracle cloud data intelligence platform. Especially, and 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. Now, this used to be back in the day. This started as a project called Cerner healthy intent. But we've fundamentally, we've rewritten Cerner healthy intent. We rebuilt it since we bought Cerner, uh, and it's become the Oracle Public Health Data intelligence platform. And we unified, uh, 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. Uh, the when you do that, when you take all of this Health Data and put it in one place, uh, you get enormous benefits. Suddenly, the first benefit is when you go to train AI models, you have

a thousand times more data than you used to have. You have all the data nor often for a drug. The best data you had you get obtain during the clinical trial. But once the clinical trial is over, they don't gather nearly as much data after. Tri after the trial is over. This is like a clinical trial that goes on forever, a clinical trial for everything that goes on forever. We keep collecting all of this data, and then we use that data to train models. Uh, and that enables the answers we get from those models enable detail personalized medicine. Now, one of the most common prescriptions, uh, in the world, uh, is a drug called EST Statin, which, uh, lowers cholesterol, uh, and there are a number of statins. And I was, you know, I happened to be on on a Statin called Crestor, and I talked to my doctor, who's a very good doctor, very fancy doctor, and a molecular biologist, and said, well, how did you decide Crestor versus and this drug or that drug or that drug? They said, well

you know you're a guy. You're, you know, um, you're of European descent. We, we really, truth is, we don't have a lot of data. It was kind of an educated guess, if you will. We thought, you know, given your family history and what we know about you, you, we thought that Prestor was a pretty good guess. Well, wouldn't you like, wouldn't I like to know, given my genomics, given everything, everything, you know about me? If you had all of my health records, you could tell me exactly what the best statin for me was, and the best statin for everybody, but not just statin, a variety of different drugs. You made, you just made better choices based on who I was and what you knew about me. You can do that. You can do that if you have this national population-scale data on public health, and you collect all the laboratory information and all the diagnostic information. You have this, this wealth of data that will help doctors, uh, make much better decisions of what therapeutics to give.

you uh and that will deliver better outcomes at a much lower cost by the way better outcomes anything keeps you out of the hospital is lower cost so strangely enough uh you know Everyone likes better outcomes the um the patient and the payer that works out for everybody uh mistakes mistakes are very costly in terms of human suffering and sometimes human life and and and and money it just bad so with with this uh new Oracle Public Health Data intelligence platform we have this fabulous training data and we can do personalized Med uh you personalized treatment get better outcomes for millions of patients I'll say it again I'm not sure there's anything we're working on here at Oracle that's more important than this uh TK and is having a session on this and I I I invite all of you who are interested in in healthare please don't miss that session okay so other things that we're we're doing um uh one of the most important things if you're training AI models is having the information having

having the data to train the AI models, and you'll be shocked that some data still isn't digitized. Some medical data isn't digitized. The quantity of medical image data, whether they're MRIs or scan or these fancy tube SE seen scanners, even sonograms aren't saved. Biopsies sit on glass slides and drawers; they're not digitized by and large. I mean a small number of digitize, but the vast majority of them are not. We don't capture this data and digitize it and vectorize it for training data, and we should. We're in the process of doing that. The cner, the cner carare, what started as cner carware again is now become a huge iot project for here at Oracle where the patient monitoring that goes on in the hospital to see you know and they keep track of your blood pressure obviously and your

temperature, and your uh, how much oxygen is in your blood, all of that because you know if you drop below a certain threshold, an alarm goes off in a nurse's station, and someone comes in to see you, but that's not recorded, that's not permanently recorded, uh, we like to start permanently recording all of that, so we'd like to make sure that the systems in the hospital, the patient monitoring systems, records all of this data so we can use it to train, uh, train systems, uh, uh, Diagnostics, uh, is said the captures the images are so large they sometimes, they're not captured at all if, if you go in, uh, and, uh, you're pregnant, and you get a sonogram, uh, that sonogram is not stored digitally, there is usually a technician or a physician that reads, looks at the sonogram, and actually measures the skull of the fetus and the length of the spine, and looks for problems like the umbilical wrapped around the, uh, that's all done looking at this sonogram image, a human being looking

at the image having a computer assisting with that uh would be a big help. Uh, for the physician or the technician recording that would be a big help if if you miss something and you go back and find go back and find it again. Uh, you shouldn't be measuring stuff on a screen. Uh, the computer should provide all of that. Uh, plus a lot of these home devices and wearables give us the opportunity to collect even more data, but all of that needs to be integrated and used as training data, and that's a project we're working on. Again, we have Partners in that effort, and we, you know, will be announcing them. The um, we capture other data. It's really interesting inside of a hospital. One of the biggest problems is their inventory. Uh, you're having a certain kind of surgery on Monday, and the surgical nurse he may on Friday run around in a variety of places and collect all of the things that you need transagamic acid to stop bleeding, the doctor's, the doctor's favorite.

gloves. You know, for operating the doctor's favorite scalpel, people have personal preferences. All of this, take it all out of inventory and hide it above in a ceiling tile, outside the operating room. So, when Monday comes, they'll have everything that they need. They withdraw it from inventory that way. They don't know what their current inventory is. They don't know where things are. They have crash carts. If they need tranexamic acid very quickly, they can go to a crash cart and get it. Tranexamic acid is something that just stops bleeding very fast. Everybody knows the first thing you do is stop the bleeding. So, one of the things we're building as part of our Fusion products actually is an IoT system that keeps track of all of this stuff. It has RFID tags on all of the inventory, and we know exactly not only how much a hospital has but where everything is located if you're looking for something. So, that's all being automated by supply chain.

Cerner had a robotics group, uh, that is automating laboratories. Automating laboratories is great, but the laboratory devices should also send the results and store them directly in the electronic health record. We need to do that. All we need to do both of those. We're, you know, we're working on that and, uh, and then robots, uh, are need to build some of these medical devices. If you, uh, if you want to make them inexpensive and highly reliable. And I think we here, we have a really interesting video of, uh, the Cerner, the Cerner, uh, robotic system that automates the laboratory. I guess do I press another button to get that video to play. Okay, we'll find out. E e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e e

오라클의 미래 비전: 래리 엘리슨 기조연설 1부

안녕하세요! "미국 빅테크 필수 번역 영상" 채널에 오신 것을 환영합니다.

미국 빅테크 기업 뉴스 번역과 분석

이 채널은 구글, 애플, 아마존, 마이크로소프트 등 미국 빅테크 기업의 최신 뉴스와 기술 동향을 한국어로 번역하여 제공합니다. 글로벌 테크 기업의 혁신적인 사례와 CEO 인터뷰를 통해 최신 정보를 쉽게 접할 수 있습니다.

주요 콘텐츠:

미국 빅테크 뉴스 번역: 최신 뉴스와 동향을 신속하게 번역하여 제공합니다.
기술 혁신 사례 분석: 빅테크 기업들의 혁신적인 기술과 사례를 깊이 있게 분석합니다.
CEO 및 전문가 인터뷰 번역: 업계 리더들의 인사이트를 제공합니다.
테크 산업 트렌드: 미래 기술 트렌드와 전망을 소개합니다.
채널의 목적:
이 채널은 여러분들이 미국 빅테크 정보를 빠르고 쉽게 이해할 수 있도록 돕기 위해 만들어졌습니다. 제가 직접 선별한 유익한 콘텐츠를 통해 최신 기술 트렌드를 파악하고, 중요한 인사이트를 얻으실 수 있습니다.

왜 이 채널을 구독해야 하나요?

신뢰할 수 있는 번역과 정확한 정보: 미국 빅테크 기업의 최신 정보를 정확하게 번역하여 제공합니다.
최신 빅테크 뉴스와 분석: 최신 소식을 빠르게 업데이트합니다.
쉽게 이해할 수 있는 콘텐츠: 테크 산업에 관심 있는 누구나 이해할 수 있는 유익한 콘텐츠를 제공합니다.
최신 미국 빅테크 소식을 놓치지 않으려면 지금 구독하세요! 유익한 정보를 놓치지 않도록 좋아요와 댓글도 부탁드립니다.
