Nvidia Is Building an AI Supercomputer and Gaming Platform

0:00Today on the AI Breakdown, everything from NVIDIA's massive AI focus to our announcement. Before that, on the AI Breakdown Brief, new ChatGPT features, a lawyer gets into hot water for using ChatGPT, and AI CEOs sign a letter comparing AI to pandemics or nuclear risk. The AI Breakdown is a daily podcast and video all about the most important news and discussions in AI. Like, subscribe, and share and find out more at breakdown.network. What's going on, guys? This is the AI Breakdown Brief, all the AI headline news you need in five minutes or less. We start today with a feature that on the one hand adds a huge amount of utility to ChatGPT, and on the other hand, the fact that it didn't exist before shows just how nascent ChatGPT's UI is.

0:46The TLDR is that ChatGPT now allows you to share links out from your conversations. So instead of just having to screenshot that particularly interesting or revelatory or brilliant conversation, you can actually get a direct link to it that can be shared with friends, colleagues, collaborators, etc. Currently, this is only available for web, and it's not available for everyone, and it's not available on the iOS app. They say, however, that that will change soon. You can also share it with your name, or you can make it anonymous. For those of you who use Google Docs, the default setting is anyone with a link can view this.

1:19Importantly, the link will also not update after it's shared, so it's really a snapshot of a moment in time rather than a living document that evolves as you interact with ChatGPT further around the same question. Lastly, shared ChatGPT links will not show up in public search results. So as I said at the beginning, it's obviously a very useful feature. But again, just remarkable how nascent this UI is, given that this most simple of functionality wasn't there until seven months into the service. Staying on the ChatGPT thread, this one got a lot of attention towards the very end of last week. A New York lawyer named Stephen Schwartz, who has been practicing law for 30 years, has thrown himself at the mercy of the courts after it was revealed that basically all of the previous case material that he cited in a client's lawsuit was fake.

2:05Yes, if you thought it was just students who were reporting hallucinated chat GPT information without double checking it, you would be very, very wrong. Schwartzman was representing a client who was suing Avianca Airlines for injuries sustained in 2019 while on a flight. After Avianca asked the judge to toss out the case, Schwartz submitted a 10-page brief that cited more than a half dozen relevant court decisions. Martinez versus Delta Air, Zickerman versus Korean Airlines, and of course, Vargis versus China Southern. The problem was that these were all invented. Mr. Schwartz now says that he, quote, greatly regrets relying on ChatGPT and, quote, will never do so in the future without absolute verification of its authenticity.

2:45Welcome to the new world of AI weirdness. Next up on the AI Breakdown Brief, some amazing new medical results from Switzerland. Gert-Jan Oskopf was in a bicycle accident 12 years ago. He was paralyzed and thought he was never going to be able to walk again. Researchers, however, have used a wireless device to connect Gert-Jan's brain to his damaged spinal cord and were able to use AI to decode thoughts and translate them into spinal cord simulation. Using the device and methodology, after 12 years, he was able to take his first steps and walk upstairs. Alex AI Daily writes, This is incredible.

3:21This is why I cover AI news, and I couldn't agree more. Now moving to something much lighter, at least on the face of it. NVIDIA's Dr. Jim Phan writes, What if we set GPT-4 free in Minecraft? I'm excited to announce Voyager, the first lifelong learning agent that plays Minecraft purely in context. Voyager continuously improves itself by writing, refining, committing, and retrieving code from a skill library. The results are that Voyager rapidly becomes a seasoned explorer. In Minecraft, it obtains 3.3 times more unique items, travels 2.3 times longer distances, and unlocks key tech tree milestones up to 15.3 times faster than prior methods.

3:57So basically what we have here is an AI agent that is specifically designed to play Minecraft, but in so doing, as it's learning, it can rewrite its own code from a skills library to improve itself continuously. All of the code is open source, so people can dig into the basis of this research and extend it in different ways. Dr. Jim goes on,

4:28Now, on the one hand, this is super interesting, right? it is bringing a different level and type of functionality to autonomous AI agents. This to many is one of the major frontiers for the next set of AI developments. At the same time, not everyone is super keen on this. Eliezer Yudkowsky writes, Presented to those of you who thought there was a hard difference between agentic minds and LLMs, where you had to like deliberately train it to be an agent or something. A, they're doing it on purpose, of course, and B, they're doing it using an off-the-shelf LLM. Now, speaking of AI risk, there has obviously been a growing conversation about this.

5:03You're seeing regulatory discussions pick up. You're seeing former industry mainstays like Jeffrey Hinton leaving their lucrative positions to start warning about these risks. And of course, a couple months ago, we had that six-month pause proposal where a number of different leaders in the space asked for a pause in training models that were more advanced than OpenAI's GPT-4. Well, now AI leaders are taking a slightly different approach. Instead of racing right towards a specific action, such as a six-month pause or anything else that might come after that, a group of leaders have signed an incredibly simple one-sentence statement.

5:36The statement reads, Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks, such as pandemics and nuclear war. The signatories for this include the folks who have been sounding the alarm most recently, including Jeffrey Hinton and Yoshua Bengio, but it also includes the CEO of Google's DeepMind, the CEO of OpenAI, the CEO of Anthropic, with dozens and dozens of other researchers, industry professionals, and others. Even Grimes, who's just about as enthusiastic as anyone about the potential of AI and our cyborg future, has signed this note. Now, the reason that I think it's interesting is from a strategic perspective, it takes a lot of the parts where you get into disagreement out of the equation.

6:16It starts to construct a shared foundation of agreement from which debates can be held more productively. This is actually a classic negotiation technique. When you have two people or two positions that are on opposite sides, you stop debating the positions and instead look at the underlying agreements between the two parties. If these leaders can get the world to agree that there is enough of a risk that it should be a global priority, that's a shared foundation from which potential remediations or approaches or pauses or any other policy strategy might be able to come from. It's a much more incremental and I think in this case smarter approach.

6:52Now there are many who still think this is just about regulatory capture and my perspective is even if the signatories who have nothing to do with the open AIs and Googles of the world suggest that it's bigger than that, we still have to be careful about whether regulatory capture is the net outcome even if it's not the intention. But either way, it's an interesting development in the conversation about AI risk. All right, guys, that's it for today's AI Breakdown Brief. If you're enjoying, please like, subscribe, and share, and I will be back soon with the main AI breakdown. From new advanced chips to an AI-focused supercomputer, these are all of the important announcements from NVIDIA's recent presentation at Computex.

7:29What's going on, guys? Welcome back to the AI Breakdown. In an extensive two-hour keynote at Computex 2023 in Taiwan, CEO Jensen Huang laid out a huge number of AI initiatives. In fact, he joked, it's too much. I know it's too much. So what we're going to do today is go through the major announcements from this presentation, figure out what they mean for NVIDIA first, but then more broadly for the AI space as a whole. Now, for a little bit of context, we have to turn to last week's stock market, because even if you had never paid attention to NVIDIA before, it's likely after last week's market performance, you probably are paying attention now.

8:05The big story heading into the stock market week last week was the US debt ceiling debate. Republicans and Democrats were at loggerheads, and it seemed like even if there was an 11th hour deal in the offing, which of course there always is, to raise the U.S. debt limit, there could be serious ramifications for U.S. debt holders looking and seeing that there was a possibility that the U.S. would actively consider defaulting on its debt. Most would have expected that to be a huge drag on the market, but it was somehow not even close to the most important story. That title went to NVIDIA. When it comes to Wall Street, everything is about expectations, and last week, NVIDIA blew them out of the water.

8:39Reporting updated Q1 earnings on Wednesday, their quarterly profit came in at more than$2 billion on revenue of more than$7 billion, which was higher than analysts' expectations of around$6.5 billion. But that wasn't the real stunner. The real stunner was in current quarter numbers, where NVIDIA was projecting$11 billion of sales, which is not only a 64 % year-over-year jump, it's more than 50 % higher than the$7.2 billion that industry analysts were projecting going into that earnings call. Susquehanna wrote, it looks like the new gold rush is upon us and Nvidia is selling all the picks and shovels.

9:13To get a sense of just how out of the norm this is, Bernstein analysts wrote, we have never seen a guide like the one Nvidia just put up. Another analyst called it without precedent and by the close of trading on Friday, Nvidia's market capitalization stood at around $960 billion, spitting difference to the trillion dollar club. Now briefly taking an even farther step back, let's talk for a moment about why NVIDIA is important to the AI industry. And I pulled up Perplexity for this because you know I love asking AI about AI questions. The two of the multiple answers to why NVIDIA is important to the AI industry that Perplexity came back with that I want to call out are one, GPU technology, and three, AI hardware.

9:51Perplexity writes, NVIDIA is a leading provider of GPUs that are widely used in AI applications. GPUs are specifically designed for parallel processing, making them ideal for handling large data sets and complex computations required in AI workloads. Now, when it comes to hardware, NVIDIA's AI hardware, such as the A100 chip, is considered the workhorse for AI professionals and is used in many supercomputers and data centers. The company's hardware platform is continuously updated to provide new features and performance improvements for deep learning. Adding a little bit of color to this, the Wall Street Journal actually just ran a piece called the AI Boom Runs on Chips, but it can't get enough.

10:25It's like toilet paper during the pandemic. Startups, investors, scrounge for computational firepower. The article starts, The artificial intelligence revolution is being likened by Google's chief executive to humanity's harnessing of fire. Now if only the industry could secure the digital kindling to fuel it. A shortage of the kind of advanced chips that are the lifeblood of new generative AI systems has set off a race to lock down computing power and find workarounds. The graphics chips or GPUs used for AI are almost all made by NVIDIA, but the boom in demand for them has far outpaced supply.

10:55The situation is restricted, the processing power that cloud service providers like Amazon and Microsoft can offer to clients such as OpenAI. AI developers need the server capacity to develop and operate their increasingly complex models and help other companies build AI services. Elon Musk told the Wall Street Journal CEO Council Summit, GPUs at this point are considerably harder to get than drugs. Now to get a sense of scale, UBS analysts estimate that an earlier version of ChatGPT required about 10 ,000 graphic chips. Elon Musk has estimated that an updated version requires three to five times that.

11:26But even as people are trying to get more of the existing AI chips, NVIDIA announced a new chip, the GH200 Grace Hopper Superchip. Systems that have the GH200 Superchip are expected to start being available later this year. But that was far from NVIDIA's only announcement at Computex. So as Barron's puts it, generative artificial intelligence applications will soon receive a massive boost in computing power. This is sort of the companion announcement to go alongside the GH200 Grace Hopper Superchips. NVIDIA's new DGX supercomputer is powered by 256 of them. According to NVIDIA, the new DGX system will enable the next generation of generative AI applications thanks to its bigger memory size and larger scale model capabilities.

12:08The DGX GH200 will have nearly 500 times the memory of its current DGX A100 system. CEO Jensen Huang said that this will allow them to help expand the frontier of AI. Now when it comes to who gets access to this supercomputer first, it's basically a who's who of the contenders for the AI space. Alphabet's Google Cloud, Meta, and Microsoft will all be among the first set of companies to get access to DGX GH200. Now it's a little beyond the technical scope of this particular video, and frankly this channel, but the other part of this supercomputer announcement is also advances in how companies can wire together these computers for even more power.

12:44In the official press release, NVIDIA writes, DGX GH200 is the first supercomputer to pair with Grace Hopper's superchips with the NVIDIA NVLink switch system, a new interconnect that enables all GPUs in a DGX GH200 system to work together as one. The previous generation system only provided for eight GPUs to be combined with NVLink as one GPU without compromising performance. The DGX GH200 architecture provides 48x more NVLink bandwidth than the previous generation, delivering the power of a massive AI supercomputer with the simplicity of programming a single GPU. And if you had any doubt about not only the focus of this supercomputer, but also the way that NVIDIA is repositioning itself as the infrastructure for AI, just look at these three official quotes from these first customers in Google Cloud, Meta, and Microsoft.

13:30Mark Lohmeyer, the vice president of compute at Google Cloud says, the new EnvyLink scale and shared memory of Grace Hopper super chips address key bottlenecks in large scale AI. And we look forward to exploring its capabilities for Google Cloud and our generative AI initiatives. Alexis Borland, Vice President of Infrastructure, AI Systems, and Accelerated Platforms at Meta. As AI models grow larger, they need power infrastructure that can scale to meet increasing demands. NVIDIA's Grace Hopper Design looks to provide researchers with the ability to explore new approaches to solve their greatest challenges.

13:58Garish Bablani, Corporate VP of Azure Infrastructure at Microsoft, says, Training large AI models is traditionally a resource and time-intensive task. The potential for DGX GH200 to work with terabyte-sized databases would allow developers to conduct advanced research at a larger scale and accelerated speeds. Now for good measure, NVIDIA also announced that it was making a supercomputer of supercomputers, and here's how they describe Helios. The supercomputer will feature four DGX GH200 systems. Each will be interconnected with NVIDIA Quantum II InfiniBand networking to supercharge data throughput for training large AI models.

14:31Helios will include 1 ,024 Grace Hopper superchips and is expected to come online by the end of the year. Now still, for all this compute power, for all of these new super chips coming online, the big moment, the thing that people are grabbing onto is a gaming development. Matt Wolfe writes, Last night, Jensen Huang of NVIDIA gave his very first live keynote in four years. The most show-stopping moment from the event was when he showed off the real-time AI in video games. A human speaks, the NPC responds in real time, and the dialogue was generated with AI on the fly.

15:11Everything is real time.

15:21Hey, Jin. How are you? Unfortunately, not so good. How come? I am worried about the crime around here. It's gotten bad lately. My ramen shop got caught in the crossfire. Can I help? If you want to do something about this, I have heard rumors that the powerful crime lord Kuman Aoki is causing all sorts of chaos in the city. He may be the root of this violence. I'll talk to him. Where can I find him? I have heard he hangs out in the underground fight clubs on the city's east side. Try there. Okay, I'll go. Be careful, Kai.

16:02None of that conversation was scripted. We gave this Jin AI character a backstory. His story about his ramen shop and the story of this game. All you have to do is go up and talk to this character. And because this character has been infused with artificial intelligence and large language models, it can interact with you, understand your meaning in a really reasonable way. All of the facial animation completely done by the AI. We have made it possible for all kinds of characters to be generated. They have their own domain knowledge. You can customize it so everybody's game's different. Look how wonderfully beautiful they are and natural they are.

16:34This is the future of video games. Not only will AI contribute to the rendering and the synthesis of the environment, AI will also animate the characters. AI will be a very big part of the future of video games. Now, NPC, of course, stands for non-player character. And what people are recognizing is that whereas NPCs traditionally have maybe just a few lines of pre-programmed dialogue, the ability to integrate generative AI conversation means that the gaming experience could become all that much more immersive. Funnily enough, although it is last year's buzzword, this type of technology actually makes the metaverse promise a lot closer to reality.

17:08Now, this is part of a larger announcement that they called Avatar Cloud Engine. As the AI kid sums up, Avatar Cloud Engine for games empowers developers to build and deploy custom voice dialogue and animation AI models on the cloud and PC, optimizing AI models for immersive responsive interactions in your software and games. Lior AlphaSignal AI writes, NVIDIA CEO Jensen Huang just announced Avatar Cloud Engine, a glimpse into what happens when gaming and AI collide. ACE is a custom generative AI model that brings intelligence to NPCs through AI-powered natural language interactions. Developers can use ACE for games to build.

17:41ACE is built on NVIDIA's Omniverse, which they call the platform for creating and operating metaverse applications. Now, some people have mentioned that the dialogue in this first demo is perhaps a little wooden, but other people are already seeing the creative possibilities. Ruprinisto writes, NVIDIA's Future of Gaming video. You talk to the characters, they talk back to you. Not like idle NPC chatter, but for the main quests themselves. Certainly a provocative idea. Imagine an Hercule Poirot game, interrogating suspects, etc. Now, for those of you who are not familiar, Hercule Poirot is Agatha Christie's famous detective, a Sherlock Holmesian type, although with his own foibles and curiosities.

18:18So he's imagining playing or taking on the role of an investigator or an interlocutor who can actually interrogate and ask questions of these formerly NPCs that now have the ability to interact in a much deeper way. I will remind you again of Jensen Huang's invocation that this is too much. It's just too much. As I point to yet another entire different section of announcements around robots. The press release reads, NVIDIA brings advanced autonomy to mobile robots with Isaac AMR. The announcement starts, As mobile robot shipments surge to meet the growing demands of industries seeking operational efficiencies, NVIDIA is launching a new platform to enable the next generation of autonomous mobile robot or AMR fleets.

18:58Isaac AMR brings advanced mapping, autonomy, and simulation to mobile robots and will soon be available for early customers. Isaac AMR is a platform to simulate, validate, deploy, optimize, and manage fleets of autonomous mobile robots. It includes edge-to-cloud software services, computing, and a set of reference sensors and robot hardware to accelerate development and deployment of AMRs, reducing costs and time to market. Now, these autonomous robots are built on something that NVIDIA calls their Nova Oran reference architecture. Nova Oran is the brains and eyes of Isaac AMR. It integrates multiple sensors, including stereo cameras, fisheye cameras, 2D and 3D LIDARs with the powerful NVIDIA Jensen AGX Oran system on module.

19:34Another benefit, they say, is that Isaac AMR accelerates mapping and semantic understanding of large environments. This helps accelerate robot mapping of large facilities from weeks to days, offering centimeter-level accuracy without the need for a highly skilled team of technicians. So the type of use case they're imagining is someone places an order for something online, the AMR is deployed in a warehouse, finds the right object, and then can get it to where it needs to go. Now showing just how many directions NVIDIA is running at once, another big announcement from this event was that they are partnering with the world's largest advertising agency to build a generative AI-enabled content engine for digital advertising.

20:08from WPP's announcement article. NVIDIA and WPP today announced they are developing a content engine that harnesses NVIDIA Omniverse and AI to enable creative teams to produce high-quality commercial content faster, more efficiently, and at scale, while staying fully aligned with a client's brand. So basically, this is a system that connects 3D design, resource libraries from companies like Adobe and Getty, and allows WPP's designers to create a huge additional volume of on-brand work. In the Computex keynote, Huang again said, The world's industries, including the$700 billion digital advertising industry, are racing to realize the benefits of AI.

20:45With Omniverse Cloud and generative AI tools, WPP is giving brands the ability to build and deploy product experiences and compelling content at a level of realism and scale never before possible. There are somehow even more announcements to this. I will include links to even more comprehensive looks at it. But the long and the short of it is that NVIDIA sees an opportunity, like just about no company in history to put themselves at the center of an industry that is changing everything around them. With all that in mind, it seems like a$1 trillion market cap is just around the corner. That's it for today's AI Breakdown.

21:16If you're enjoying, please like, subscribe, and share. Check out the podcast in the newsletter. And until next time, peace.

21:33Thank you.