AI transcript
0:00:04 Hey, welcome to the Next Wave Podcast.
0:00:08 I’m Matt Wolfe, and in today’s episode, we’re talking about AI agents.
0:00:15 I had the opportunity to go spend a few days with NVIDIA out at the NVIDIA GTC conference.
0:00:20 And in this episode, I’m going to deep dive with Amanda Saunders from NVIDIA.
0:00:26 She’s been at the heart of the AI agent revolution that’s happening right now while working with
0:00:30 NVIDIA, the company that’s enabling the AI revolution.
0:00:36 In this episode, we’ll unpack exactly what agentic AI is and how it’s reshaping industries
0:00:41 from healthcare and telecom to sports coaching, as well as why that’s just scratching the surface
0:00:42 of what’s possible.
0:00:48 You’ll also discover NVIDIA’s secret blueprint for easily building powerful AI agents, as
0:00:54 well as get into some of the fears that people have around AI agents, like is an AI agent going
0:00:56 to take your job and replace you?
0:00:57 Yeah, we’re going to go there.
0:01:01 And after the interview, I’m going to share some cool clips from NVIDIA’s GTC, where I
0:01:09 got Bob Petty, also from NVIDIA, to break down exactly what NVIDIA’s DGX Spark is and how
0:01:14 they’re going to be putting AI supercomputers in normal people’s homes.
0:01:18 You can have an AI supercomputer in your home by this time next year.
0:01:20 We’re going to get into that in today’s episode.
0:01:25 So without further ado, here’s my discussion with Amanda Saunders, followed by my tour of the
0:01:27 EGX Spark with Bob Petty.
0:01:28 Enjoy.
0:01:33 Marketing in 2025 is wild.
0:01:36 Savvy customers spot fake messaging instantly.
0:01:42 Privacy changes make ad targeting a nightmare, and everyone needs more content than ever.
0:01:46 That’s why you have to have HubSpot’s new Marketing Trends Report.
0:01:51 It doesn’t just show you what’s changing, it shows you exactly how to deal with it.
0:01:55 Everything’s backed by research, but focused on marketing plays you can use tomorrow.
0:01:58 Ready to turn marketing hurdles into results?
0:02:02 Go to click HubSpot.com slash marketing to download it for free.
0:02:07 Yeah, so my name’s Amanda Saunders.
0:02:11 I’m the director of enterprise AI software here at NVIDIA.
0:02:16 So you know a lot of people think about NVIDIA and all the amazing GPUs, CPUs, and DPUs that
0:02:18 we make that go into these powerful systems.
0:02:21 Well, actually, we have a lot of software that runs on that.
0:02:26 So me and my team, we focus on how do we tell that story and bring it to customers.
0:02:31 I’ve been at NVIDIA 10 years now, and I’ve covered everything from, you know, the graphic
0:02:34 side of our business into the data science.
0:02:37 And now, of course, the hottest topic, generative AI.
0:02:38 Very cool.
0:02:41 Let’s talk about agentic AI, because that’s not a hot topic.
0:02:45 It seems like in 2025, everybody’s saying this is the year of agentic AI.
0:02:48 How would you define agentic AI?
0:02:49 Absolutely.
0:02:52 So agents are helpful software tools.
0:02:56 They are digital employees that help us augment our work.
0:03:01 And what’s really important about agents and agentic AI is that it can perceive.
0:03:02 So it sees the world around it.
0:03:04 It sees the data that it has access to.
0:03:06 It takes that information.
0:03:08 It reasons about that information.
0:03:09 It thinks about it.
0:03:13 And then it can actually make actions based on that data.
0:03:16 So it can perceive, it can reason, and it can act.
0:03:20 And that’s what really sort of sets it apart from generative AI that we had in the past,
0:03:26 is being able to take those actions, whether that action is alerting a human or actually
0:03:30 actively using a tool and making something happen.
0:03:34 And so, yeah, I think they’re really exciting because they’re able to solve complex problems
0:03:37 that we’ve never been able to do with software before.
0:03:37 Very cool.
0:03:41 So what do you think makes agentic AI powerful right now?
0:03:42 Why not a year ago?
0:03:43 Why not two years ago?
0:03:46 Why is now the time everybody’s talking about it?
0:03:48 You know, I think this has been a journey that we’ve been on.
0:03:51 You know, it started with having to have the accelerated computing.
0:03:53 You had to have the computing power to do that.
0:03:56 And that’s a problem that NVIDIA has been working on for 30 years.
0:03:59 And then finally, we’ve got to the place where we have the computing systems.
0:04:01 We also then needed the software systems.
0:04:05 And so it started with the open models that became available.
0:04:09 I think LAMA was a huge advancement and step forward.
0:04:12 And more and more models have come out recently.
0:04:16 Reasoning models, in fact, are one of the big pieces that have sort of driven that
0:04:18 and sort of added to that.
0:04:22 And then just the full ecosystem of tools and things that are required.
0:04:28 I think AI is one of those really interesting fields where the more people use it, the more
0:04:31 they want to use more of it and they want to do more.
0:04:35 And so it’s really spurring this incredibly fast growth that we’re seeing, which is, you
0:04:36 know, breakneck speed.
0:04:42 But that’s what’s really driving us from, you know, the early onsets of generative AI,
0:04:46 where you’re just spitting out answers into these worlds now where they’re actually thinking
0:04:50 things through and making really smart actions, which is just sort of cool.
0:04:50 Yeah.
0:04:55 Or in my case, I got into AI and now I just make content about it because I’m obsessed with it.
0:04:56 It’s all I ever want to talk about.
0:05:00 Because yeah, once you get into it, you really, that’s all you want to do is you figure out
0:05:01 what it can do.
0:05:05 You learn where it’s, you know, boundaries are, and then you want to do more and more and
0:05:05 more.
0:05:10 And as it continues to, you know, get better, you can do more and more things with it.
0:05:10 So cool.
0:05:13 Well, can we walk through like a agentic workflow?
0:05:17 Like give some examples of like, this is the types of things an agent can do.
0:05:18 Yeah, absolutely.
0:05:23 I think, you know, some of the most basic things that we see from agents are really about
0:05:24 being able to talk to data.
0:05:28 I think that’s probably the first thing that we see most people try to do is they have their
0:05:34 own personal data, they connect an LLM to that data, and then they’re able to ask questions.
0:05:38 And I think, you know, more recently, they’re able to ask more and deeper questions.
0:05:40 Deep research has been a big topic.
0:05:42 It’s a big topic here at GTC.
0:05:48 And the more information you can get out of that data and the more quickly you can query that,
0:05:51 that’s a pretty standard agent that we would see out there.
0:05:54 Now, then they start to get a lot more complex.
0:05:59 I think there’s some really cool stuff going on in the telecom space where they’re actually
0:06:05 using agents to improve the network so they can actually predict when there might be network
0:06:12 outages coming and they can make recommendations to the human employees who, you know, maybe you
0:06:17 want to make these changes because maybe there’s a large show in town like GTC and there’s going
0:06:20 to be a lot of traffic on that network.
0:06:25 Here are the recommended changes so that you can actually, you know, still provide the service.
0:06:27 Is that kind of thing happening at GTC right now?
0:06:28 Are they using that?
0:06:29 I wish they would.
0:06:33 I think, I think the agents are just starting to be developed.
0:06:37 So maybe next year we’ll actually see that, but it’s definitely something that we’re going
0:06:37 to see more of.
0:06:38 Yeah.
0:06:38 But yeah.
0:06:41 So those I think are, you know, some varying levels of examples.
0:06:45 And then there’s obviously hundreds more in the healthcare space, you know, helping nurses
0:06:50 and doctors become more efficient because we know they need the help to, you know, everyday
0:06:51 people like you and I.
0:06:56 I mean, I use it for almost everything I do, whether it’s in my personal life or my
0:06:56 work life.
0:06:59 Are there any like agents that might surprise people?
0:07:04 Anything that’s like really sort of like interesting that, oh, I wouldn’t have thought people would
0:07:05 be using agents in that area.
0:07:07 It’s an interesting question.
0:07:12 I mean, the most interesting one for me on the agent side is really, as you started to
0:07:16 get into video and other types of data sources, like text, I think people have gotten the handle
0:07:21 on, but adding sort of video understanding and things coming through it.
0:07:25 Actually, one of the really cool use cases that we recently did was Jensen got to throw
0:07:27 out the first pitch at a baseball game.
0:07:28 I saw the clip of that.
0:07:36 And we used a video agent to be able to critique his performance, which I just think is really
0:07:36 cool.
0:07:41 So you can imagine for, you know, athletes, whether they’re amateurs or professionals, being
0:07:43 able to use that is really helpful.
0:07:43 Yeah.
0:07:44 Yeah.
0:07:46 I can see that like golfers and stuff like that.
0:07:47 Analyze my swing.
0:07:48 Tell me what I’m doing wrong.
0:07:48 Stuff like that.
0:07:53 And it’s probably too much for the AI on my swing, but you know, it still would be helpful
0:07:54 to know.
0:07:54 Yeah.
0:07:54 Yeah.
0:07:56 Well, let’s, let’s talk about the NVIDIA blueprints.
0:08:00 Because when I was at CES, that was a big topic during CES was the NVIDIA blueprints.
0:08:03 And to me, that’s sort of like the beginnings of agents, right?
0:08:06 But there’s sort of this pre-built agent.
0:08:07 You could probably explain it better than I can.
0:08:09 Well, exactly.
0:08:12 We tried to name them blueprints so that people would get the idea that these are reference
0:08:13 architectures.
0:08:18 And what they do is they take the different building blocks that NVIDIA is offering to
0:08:22 make these agents, and it helps you with a recipe on how to put them together.
0:08:25 So it starts out with our NIM microservices.
0:08:29 And so these are packaged, containerized, optimized models.
0:08:30 Right.
0:08:34 So all the leading open models that are out there in the market, we take them, we containerize
0:08:39 them, we add a standard API call to them so that people, you know, developers out there
0:08:40 can start building.
0:08:42 And we package those up.
0:08:43 So now you’ve got the model.
0:08:45 And now you need to connect that model.
0:08:46 Right.
0:08:46 It’s right.
0:08:47 A model and it’s only, only does so much.
0:08:48 So that’s…
0:08:54 So like a NIM would be like, you’ve got a video NIM that this one NIM can produce AI-generated
0:08:55 video.
0:08:57 This NIM could produce text-to-speech.
0:08:59 This NIM could do speech-to-text.
0:08:59 Exactly.
0:09:00 Right.
0:09:00 Okay.
0:09:00 Exactly.
0:09:05 So we actually have a hundred NIM that are now available that you can download and use.
0:09:10 And they do, there are 40 different sort of domains or modalities that these different
0:09:11 NIM can do.
0:09:12 So yeah.
0:09:14 So you can take a bunch of them, piece them together.
0:09:17 And so when we built those, everyone was like, this is great.
0:09:18 We need the models.
0:09:22 We need them to run really well on, you know, our NVIDIA GPUs.
0:09:24 But then they said, well, but how do we build them into the next thing?
0:09:26 And that’s where Blueprints came from.
0:09:26 Right.
0:09:31 So they are the recipe that allows you to take the NIM, start with this recipe, piece this
0:09:32 together, and you’ll get to an agent.
0:09:36 And then what’s even better is you can then customize them.
0:09:38 So you can, you know, add your own data sources.
0:09:40 You can add your own pieces in.
0:09:42 You can combine Blueprints together.
0:09:47 So maybe you started with a chatbot and you wanted to add a digital human on the front end.
0:09:49 We have a Blueprint for both of those.
0:09:52 You piece them together and all of a sudden you have a digital avatar.
0:09:57 Who can, you know, talk to you with a full, you know, face and expressions and natural language.
0:09:58 Probably even your voice if he wanted to.
0:09:59 Yeah.
0:09:59 Absolutely.
0:10:04 And you can, you can start with pieces from NVIDIA or you can start with pieces from the
0:10:04 ecosystem.
0:10:07 We try to make it really easy to piece it all together.
0:10:08 Super cool.
0:10:10 Let’s talk a little bit about like security.
0:10:15 I know when it comes to AI agents, they can be used for both good and bad.
0:10:19 What kind of things can we do to sort of protect and secure and make sure that people aren’t
0:10:23 going and using these agents for, you know, bad actor type stuff?
0:10:23 Absolutely.
0:10:30 I think, you know, what’s really cool about AI is where it raises concerns or security challenges
0:10:32 and it can actually also help answer them.
0:10:36 So there’s a Blueprint out there for container security that does a lot of those pieces.
0:10:43 But we also have other applications that can actually track activities and make alerts based
0:10:45 on detecting some anomaly, right?
0:10:49 So that’s tends to be how you recognize that something’s doing something it maybe wasn’t
0:10:50 supposed to.
0:10:54 And AIs are just really good at that because they can, you know, consume a lot of information.
0:11:00 So I think where, you know, AI potentially can raise some concerns, it’s also in some ways
0:11:03 the answer to addressing some of those concerns, which I think is really helpful.
0:11:03 Right.
0:11:09 So like the sort of like sci-fi movie scenario of like the AI rising up against us.
0:11:14 I mean, is this something that people should be concerned about or is this something that
0:11:16 you feel there’s pretty good guardrails in place for?
0:11:20 I think the builders of these apps need to make sure those guardrails are in place.
0:11:27 But I think, yes, in general, I think the tools exist and then it’s just about us as the application
0:11:29 builders being smart about how we deploy them.
0:11:30 Right.
0:11:33 You know, in general, I think agents are really powerful.
0:11:38 They’re incredible tools that help humans, but on their own, they’re not about to sort
0:11:39 of go off the rails.
0:11:41 It really takes a human to take it there.
0:11:43 So I think, again, it is a tool.
0:11:46 It is something that can, you know, work alongside you.
0:11:50 But I don’t see the rise of the machines quite yet.
0:11:55 So along similar lines, I think one of the fears a lot of people have is like, is an agent
0:11:56 going to take my job?
0:12:00 So, you know, what are your thoughts in that sort of realm?
0:12:06 I mean, for me personally, I see all of AI as like a tool, something that makes me a lot
0:12:09 more efficient and makes it so I can accomplish more in a day.
0:12:11 But I’m curious, like, what’s your take?
0:12:13 Do you think that, you know, people need to be concerned about that?
0:12:18 I think the best thing I’ve ever heard is an agent’s not going to take your job, but somebody
0:12:19 using an agent might.
0:12:21 And that’s always the thing that I think gets framed.
0:12:23 No, I think exactly that.
0:12:27 Like we as humans have the capacity to do so much more.
0:12:29 There’s just often so many hours in the day.
0:12:29 Right.
0:12:35 And so we don’t, you know, accomplish, you know, with agents, the work that we were trying
0:12:36 to do and then just stop.
0:12:36 Right.
0:12:39 No, it allows us to go and do more and try more.
0:12:42 And so I think, you know, for me, it’s not about replacing your job.
0:12:47 It’s about making you more effective at your job and being able to do more and, you know,
0:12:48 be more powerful.
0:12:53 I think, you know, healthcare for me is one of those industries that this is driven out there.
0:12:58 There’s actually a great partner of ours in the mental health space and they’re using
0:13:03 agents to free up therapist time from all the administrative work that they have to do.
0:13:05 We know their jobs are hard enough as it is.
0:13:11 If we can free that time up, it allows them to spend more time with their patients, actually
0:13:17 deeply understanding and working with them as opposed to worrying about calendars and taking
0:13:17 notes and scheduling.
0:13:22 So I think those are great examples where it’s, no, it’s something there that’s going
0:13:23 to help your job.
0:13:24 It’s not going to take your job.
0:13:24 Yeah.
0:13:28 I bumped into somebody on the street while I was walking yesterday and they had a very similar
0:13:29 concept for their company.
0:13:34 They work with therapists and they create chatbots specifically for the therapist so that their
0:13:38 customers can go and have the initial conversation with the chatbot.
0:13:40 But then the conversations get passed along to the therapist.
0:13:45 The therapist can kind of, you know, detect when some bigger issue that they need to address
0:13:47 comes up, but it’s not replacing the therapist.
0:13:50 It’s just, hey, I can now handle more patients, you know.
0:13:53 And give them more of my dedicated time.
0:13:53 Right.
0:13:57 And I think there’s something actually interesting as you start to see chatbots and particularly
0:14:03 those with digital avatars is humans will open up in some ways to one of these, you know,
0:14:07 chatbots or these avatars in ways that they may not necessarily feel comfortable opening up.
0:14:11 So I think it’s just giving us new avenues to do things that we were already doing anyway.
0:14:13 And I think that’s pretty cool.
0:14:13 Yeah.
0:14:17 This is sort of switching gears a little bit, but do you think there’s like a compute bottleneck
0:14:17 for agents?
0:14:21 Are we going to be able to scale agents and get to agents as fast as people would want
0:14:21 to?
0:14:25 There was a narrative, I feel like the narrative sort of faded a little bit, but there was
0:14:28 a narrative maybe six months ago that there’s a wall and AI is hitting a wall.
0:14:29 Do you think there’s a wall?
0:14:31 Do you think we’re going to hit some compute bottlenecks?
0:14:36 I think, you know, this is our life’s work at NVIDIA is to make sure that we have the compute
0:14:39 to, you know, drive the world’s agents.
0:14:44 And so I think, you know, a lot of the announcements we talked about today, both on the hardware and
0:14:49 on the software side are really focused on making these as efficient as possible.
0:14:55 And I think what’s really cool is as you’re looking at this AI space, when it first comes
0:14:58 out, it’s usually big, it’s boxy, it’s maybe not the most efficient.
0:15:03 And then over time, that efficiency comes in and allows us to do the next big leap.
0:15:06 And again, that starts out, you know, big and heavy.
0:15:09 And again, gets more and more efficient over time.
0:15:13 So I think I’m not seeing a bottleneck, but I do know it’s an area we need to continue
0:15:18 to drive because I do think the compute requirements are going to continue to grow.
0:15:18 Cool.
0:15:21 Well, so let’s extrapolate out a little bit.
0:15:24 So if you’re looking like five, 10 years down the road, what do you sort of envision with
0:15:25 AI agents?
0:15:26 Where do you think this is all headed?
0:15:29 And I know that’s hard because it’s sort of exponential technology.
0:15:31 It’s really hard for humans to grasp back credentials.
0:15:32 It’s incredibly hard.
0:15:36 I mean, I can barely keep six months out, you know, and new things keep popping up.
0:15:41 I mean, I think certainly the biggest things that we’re going to see are teams of agents.
0:15:44 And that’s going to be, you know, that we’re already seeing it today.
0:15:46 It’s starting now and it’s going to continue to grow.
0:15:52 And so, you know, the way that I see it is, you know, when we first got agents and we first
0:15:57 got, you know, chatbots and things, you could ask a question and it would respond to a question.
0:16:00 And then you could ask it to do a function and it could now do that function.
0:16:02 And then you can ask it to do a bigger job.
0:16:05 And these agents are just getting more and more sophisticated.
0:16:10 So five to 10 years out, you know, we may be able to give something, you know, as simple
0:16:14 as, you know, design me my, you know, retirement home.
0:16:18 And it will come back with all of the pieces involved in that without a human ever having
0:16:19 to do another prompt to follow it up.
0:16:21 And I think that could be pretty cool.
0:16:24 So that might even be just two or three years.
0:16:27 But yes, I certainly see that.
0:16:28 I think that’s where we’re going.
0:16:29 So you mentioned teams of agents.
0:16:30 That’s really fascinating to me.
0:16:35 Like, do you have any examples of like what sort of agents would team up to work together
0:16:36 and what sort of tasks will accomplish?
0:16:37 Yeah, absolutely.
0:16:38 So I work in marketing, right?
0:16:44 And so a lot of my job is about doing messaging, writing blogs, getting imagery, building
0:16:45 demo videos, things like that.
0:16:50 And today, you know, you start with one, you know, app that can help you write.
0:16:52 And then you go to another app that helps you build images.
0:16:55 And then you go to another app that helps you write the code for the website.
0:16:56 And then, you know, right.
0:16:58 And they’re all separate.
0:17:00 And each one of them returns to me.
0:17:03 And I, you know, move to the next step.
0:17:09 I think in the future, what we’re going to see is those agents will all be connected through
0:17:09 function calls.
0:17:14 And we actually, uh, at GTC announced a blueprint that’s going to help us do this.
0:17:15 It’s called IQ.
0:17:21 And so by connecting all these together, again, as a human, I’ll be able to start putting
0:17:23 my overall requests.
0:17:27 I need to build a new website for a new announcement that’s coming out and it will be able to do
0:17:29 all those functions together.
0:17:34 And I think what’s really cool about this is it’s going to allow us to design agents that
0:17:40 solve specific tasks, but by combining them together in that sort of that composable way,
0:17:43 they’ll be able to do bigger and bigger job functions.
0:17:43 Very cool.
0:17:49 So like what sort of bigger world plot problems do you see like AI and AI agents solving for
0:17:49 us?
0:17:54 I mean, I think networking is a really interesting one because anything that has that much data
0:18:01 that humans can’t solve, I think is, is amazing, you know, digital twins and simulation of those
0:18:06 types of environments, whether it’s, you know, the climate and weather, whether it’s, you know,
0:18:08 the businesses and things that we run.
0:18:13 I think all of these are areas where agents are just going to add to the ability to work with
0:18:14 these.
0:18:16 So, yeah, certainly I think those are big overall.
0:18:22 And then I think businesses, of course, it’s going to be about, you know, providing those
0:18:25 tools so that their employees are just that much more effective.
0:18:25 Yeah.
0:18:30 And that’s, that’s not a big world problem, but it’s a very common serious world problem.
0:18:30 Yeah.
0:18:34 One of the things that I think was really fascinating and I’ve heard, I’ve, I’ve made jokes about
0:18:38 how I feel like I’m on like the Jensen tour because I’ve actually seen his last like five
0:18:38 keynotes.
0:18:41 But one of the things that I’ve, I’ve seen him talk about that’s really fascinating to
0:18:47 me is the earth too, where it’s got the, they basically couldn’t map out the weather patterns
0:18:49 and figure out weather events a lot earlier.
0:18:54 So I’m really excited to see the sort of overlap of like the earth to concept and agents and solving
0:18:57 some of the more like bigger climate type issues as well.
0:18:58 A hundred percent.
0:19:03 I mean, I think one of the things when, when we introduced earth to is this idea that we,
0:19:08 we think a lot about these problems, we have conversations about them, but it’s really hard
0:19:13 as humans to sort of visualize what, you know, some of the changes that we make in government
0:19:16 and things like that are going to do in the next five to 10 years.
0:19:18 We’re very immediate creatures, right?
0:19:19 So that’s where we’re focused.
0:19:23 But I think, you know, imagine being able to go to earth too and through an agent say, Hey,
0:19:26 you know, if we made these changes, what would happen?
0:19:32 Or if we, you know, took these steps, how could that affect the world and what more can we
0:19:32 do?
0:19:37 And maybe there are things that we’re not even recognizing that the agent could recommend.
0:19:41 Because I think that’s, what’s really cool about these agents is they’re not humans.
0:19:43 They don’t think the way we do.
0:19:48 And so by giving them all that data and giving them an earth simulation, they might be able
0:19:49 to uncover things that we’ve never thought of.
0:19:50 Yeah.
0:19:50 Yeah.
0:19:51 I think that’s pretty cool.
0:19:51 Yeah.
0:19:52 I think that’s amazing.
0:19:57 What are some of the things in the AI world, you know, agents or otherwise that have you personally
0:19:58 really excited?
0:19:59 Like what sort of stuff do you use?
0:20:00 What do you play with?
0:20:02 Like, what’s your AI stack that you use?
0:20:05 I mean, I use as many as I can get my hands on.
0:20:08 We have a lot of them in NVIDIA.
0:20:12 I mean, one of my favorites, and I know Jensen talked about it a little bit on stage, is perplexity.
0:20:13 Yeah.
0:20:13 Oh, I love perplexity.
0:20:19 I love perplexity because I think it changes the dynamic of humans and how we interact.
0:20:25 Rather than searching for information and having to spend the thought process on that, it’s really
0:20:27 about who can ask the best questions.
0:20:33 And I think that’s a really powerful change and, you know, power dynamic that perplexity
0:20:34 is given to the users.
0:20:36 Now, if you can ask great questions, you can find great information.
0:20:38 So I love that tool.
0:20:44 We have AI agents in our company that help us with everything from, you know, our benefits
0:20:48 and understanding how to make the right decisions, you know, for each employee, which I think is
0:20:51 pretty cool, into, you know, how we do our jobs.
0:20:56 Whether that’s video creation, image creation, certainly content, which you have to write
0:20:57 a lot of.
0:20:58 So, yeah.
0:20:59 So all of those, I think, are great.
0:21:04 And then, you know, I take them into my personal life in terms of organizing and planning.
0:21:09 I’m very organized at work and I don’t always save a lot of that organization for my personal
0:21:10 life.
0:21:13 And so I can hand it off to, you know, any sorts of chatbots.
0:21:16 I think ChatTPT is excellent for this.
0:21:17 It’s just pretty cool.
0:21:18 And lately, deep research.
0:21:20 So I’ve been using a lot of that.
0:21:25 And again, we just introduced a blueprint that’s going to allow deep research on your
0:21:26 own personal things.
0:21:28 And so I can imagine that’s going to be pretty powerful.
0:21:33 So for people that want to sort of stay in the loop on AI that are curious, maybe they’re
0:21:33 worried about it.
0:21:37 Like, what sort of advice would you give them to sort of stay on top of things?
0:21:41 My best advice on AI is use it.
0:21:46 It sounds really obvious, but I think, one, I think it actually alleviates a lot of concerns
0:21:48 when you understand how the technology works.
0:21:54 And by using it, you can see what works today, where the limitations are, and how it kind
0:21:55 of functions.
0:22:02 And I think it lets people see it as that tool versus that, you know, something that might
0:22:02 be scary.
0:22:06 So that’s really the first piece of advice.
0:22:07 I think that’s really important.
0:22:12 And then I think it’s about, you know, trying to identify where many of those, you know,
0:22:17 concerns might be coming from, whether that’s data or security or things like that, and understand
0:22:19 how AI can also help with those.
0:22:23 And so that tends to be, you know, one of my best pieces of advice.
0:22:23 Cool.
0:22:24 Yeah.
0:22:26 Well, let’s talk about Llama 3 and Nemo Tron.
0:22:27 Yeah.
0:22:28 So Llama, Nemo Tron.
0:22:29 Llama, Nemo Tron.
0:22:30 Let’s talk about Llama, Nemo Tron.
0:22:31 Absolutely.
0:22:33 So Llama, Nemo Tron was a really cool announcement.
0:22:38 So NVIDIA, we work with all of the leading model builders that are out there, including Meta,
0:22:39 who released Llama.
0:22:46 And what’s really cool about Llama is there are, I think it’s 85,000 derivatives of this model.
0:22:51 And so, of course, NVIDIA, we’ve got a lot of really smart people who know how to optimize
0:22:53 and make models more efficient.
0:23:00 And when Reasoning came out and when DeepSeek introduced this reasoning wave into the open
0:23:06 source, we sort of said, well, how can we bring this and make it really efficient for,
0:23:08 you know, people who want to deploy this on NVIDIA?
0:23:15 And so by starting with the Llama model, which is an incredible model, we brought in our expertise
0:23:20 to train it so that a model that previously couldn’t do reasoning now could actually think
0:23:21 through problems.
0:23:26 And so we found, you know, the data set to go be able to train the model and how to do
0:23:27 this new task.
0:23:32 And we trained it and then we made the data source open, which I think is really cool.
0:23:36 So if others want to do their own training or if they want to train a different model or
0:23:38 anything like that, that data source is available.
0:23:44 But it’s just teaching a model a new skill is, I think, a really powerful thing to see because
0:23:47 it shows how quickly the space is evolving.
0:23:51 So is there anything actually happening like underneath Llama or is it just you’ve got
0:23:55 the Llama model, but now there’s a new layer on top of it that knows how to think?
0:23:57 Is there did any like new training happen to the model?
0:23:58 Yes.
0:24:00 So, yeah, the model was absolutely.
0:24:01 So we post train the model.
0:24:05 So this is what a lot of companies out there are doing today is they take a base model and
0:24:07 they actually train it with new data.
0:24:11 And in this case, sometimes you’re training it so it has more information on a particular
0:24:12 topic.
0:24:17 This is particularly popular when you’ve got domains and industries that have specific languages
0:24:20 that they speak, things like maybe the finance industry.
0:24:24 But in this case, we actually were training it on a skill.
0:24:27 And that skill is to think through problems.
0:24:31 And I think this is what’s really interesting about reasoning is the way a reasoning model
0:24:36 works is it starts by thinking through the question that you’re asking.
0:24:40 And it breaks down that question into multiple steps and multiple parts.
0:24:45 And then it actually goes through and comes up with answers for each of those parts.
0:24:51 And then it checks to say, OK, now that I’ve got those answers, does this actually, you know,
0:24:53 come back with the right question?
0:24:58 And it continues to do that until it gets to this highly accurate response.
0:25:05 And so not only are reasoning models really good for improving accuracy, which we all know
0:25:07 that when an AI model is useful, it’s when it’s accurate.
0:25:08 Right.
0:25:11 But it also allows us to solve problems we can never solve.
0:25:12 Right.
0:25:15 So a great example for me on this is I love puzzles.
0:25:20 I love all sorts of puzzles, but particularly Sudoku, because I hear it’s good for your brain.
0:25:27 Sudoku is a problem that humans can solve, but actually traditional LLMs couldn’t.
0:25:34 There are actually almost 80 different decisions that go into solving a Sudoku puzzle.
0:25:39 And each one of those affects the other decision, because obviously based on the rules that you
0:25:44 have to follow, reasoning models, and Lama Nibodron is a great example of this, can solve Sudoku.
0:25:45 Ooh, interesting.
0:25:48 Where a Lama model on its own couldn’t.
0:25:48 Wow.
0:25:50 And so again, that’s the type of thing.
0:25:54 It doesn’t sound like, you know, solving Sudoku puzzles is going to change the world.
0:25:59 But when you look at a Sudoku puzzle, there’s actually a lot of things that go on in the
0:26:01 world that are related to that.
0:26:02 Yeah.
0:26:03 I have a supply chain.
0:26:07 I’ve got to ship items and get them to different stores around the country.
0:26:12 I happen to know there’s a snowstorm coming in and I need to make sure that my trucks are
0:26:14 taking the most efficient routes.
0:26:18 All of those are steps that impact the other decisions that are being made.
0:26:23 And so it’s actually quite a complex problem that relates in some ways to Sudoku.
0:26:24 Yeah.
0:26:28 Well, it’s so interesting too, because you take a normal model and it can’t tell you
0:26:32 how many R’s are in strawberry, give it a thinking model, and it will actually count and then
0:26:33 double check.
0:26:34 Did I do that right?
0:26:34 And then answer.
0:26:36 And get the right answer.
0:26:36 Exactly.
0:26:40 It’s so fascinating because it seems like such a simple problem to a human brain.
0:26:44 But then that’s, I think, the thing is we look at models and we think of them like
0:26:46 we personify them as humans and they’re not.
0:26:47 Right.
0:26:49 And so, yeah, reasoning models are really cool for that.
0:26:53 And we’ve seen a lot of those great examples that come out of what these reasoning models
0:26:53 can do.
0:26:55 And it is.
0:26:57 It’s just, I think it’s really interesting to watch.
0:26:57 Yeah.
0:27:01 It’s so cool to actually see the thought process because you’ll actually see the models think
0:27:03 through something and then go, wait, that’s probably not right.
0:27:05 Let me think that through again.
0:27:08 And you actually see that text come out of it thinking through.
0:27:09 And that to me is fascinating.
0:27:10 It’s fascinating.
0:27:15 And it’s a great example of the compute story we were talking about, which is, you know, we do
0:27:17 need more compute so it can think.
0:27:19 The more it thinks, the more compute it requires.
0:27:24 So it’s this really interesting cycle that we’re watching these things go through.
0:27:28 One of my favorite things to do with reasoning models is to ask them to describe things to
0:27:29 me like a five-year-old.
0:27:29 Yes.
0:27:32 And it will come up with a description.
0:27:35 It will check if a five-year-old would understand that description.
0:27:36 It will then make changes.
0:27:39 And it’s really interesting to see what models think five-year-olds understand.
0:27:41 So at some point, I’ll have to go test it with a real question.
0:27:43 Yeah, read this.
0:27:44 Does this make sense to you?
0:27:45 Does this actually make sense?
0:27:46 Exactly, exactly.
0:27:47 Very cool.
0:27:50 Well, along the same line, real quick, let’s talk about hallucinations.
0:27:53 Do you see like a path to zero hallucinations?
0:27:55 Do we want to get rid of hallucinations completely?
0:27:56 Like, what are your thoughts on that?
0:28:01 I think, you know, there’s certainly things we can do to reduce hallucinations.
0:28:07 And some of those are, you know, as simple as putting a guardrail in place that says, if you’re
0:28:12 not 100% confident in the answer or 99% confident in the answer, don’t answer the question.
0:28:12 Right.
0:28:16 That sort of, it stops the model from doing things that it shouldn’t.
0:28:22 And Nemo guardrails, which is one of the products that NVIDIA offers, helps with building those.
0:28:26 But to your point about, do we want to stop hallucinations altogether?
0:28:31 Of course, we want it to give the right answers, but in being creative, we’re asking it to come
0:28:32 up with new things.
0:28:32 Right.
0:28:37 So, it’s, you have to be able to distinguish between a hallucination and a creative generative
0:28:37 response.
0:28:38 Right, right.
0:28:41 So, I think that’s where this balance plays.
0:28:47 So, there are absolutely steps that can be taken, and depending on how targeted and focused
0:28:51 you want the model to be, you can put more and more of those sort of guardrails or those
0:28:53 policies in place that will keep it from hallucinating.
0:28:54 Right.
0:28:57 Yeah, because I think in a lot of scenarios, hallucinations are a feature, not a bug, right?
0:29:01 If you wanted to write a short story for you, you wanted to hallucinate that short
0:29:01 story for you.
0:29:02 Exactly.
0:29:06 So, I think that’s where we have to understand what’s the hallucination versus what’s the model
0:29:07 doing what it’s supposed to do.
0:29:08 Right.
0:29:10 And again, and that’s where it becomes, what’s the use case?
0:29:12 What are you trying to have it do?
0:29:17 And really think those through, and then find the right tools from NVIDIA or others to be
0:29:19 able to actually, you know, go and do that.
0:29:20 Very cool.
0:29:24 Yeah, so if somebody wanted to get started with agentic AI, they want to start playing with
0:29:26 agents and testing the waters, what do they do?
0:29:27 What are our steps?
0:29:31 Well, so, from NVIDIA, we have something called build.nvidia.com.
0:29:33 We made the URL super easy.
0:29:36 If you’re trying to build something, we have a one-stop shop for you.
0:29:39 It’s got all the models on there so that you can test them out.
0:29:43 Whether it’s the new Llama Nemetron model with reasoning, you can actually turn reasoning on
0:29:43 and off.
0:29:45 It also has all the blueprints.
0:29:47 So, you can actually test them and experiment from them.
0:29:52 And then from there, there are also steps to go deploy them, test them out, and build them
0:29:53 yourself.
0:29:55 So, I think that’s a great starting point.
0:29:56 Very cool.
0:30:00 And for the more, like, technical people that maybe are trying to develop something, like,
0:30:02 is there a place they can go play with the APIs?
0:30:04 Like, what do we do there?
0:30:05 Build.nvidia.com.
0:30:05 Same place.
0:30:06 Same place.
0:30:11 It’s literally, whether you’re, you know, an enthusiast, whether you’re a developer, whether
0:30:16 you’re actually trying to, you know, build something to put in production, this is the one-stop
0:30:18 shop because everything’s on there.
0:30:20 You can take it as far as you want.
0:30:21 You can play around with the UI.
0:30:23 You can play around with the APIs.
0:30:28 You can actually download and deploy these models on any, you know, NVIDIA hardware.
0:30:30 It’s your one-stop shop for everything.
0:30:30 Yeah.
0:30:32 Actually, that brings me to another question.
0:30:35 Does this stuff work on older NVIDIA hardware?
0:30:40 If you have a, you know, a 3080 or a 4070, can I use this stuff on those as well?
0:30:41 Absolutely.
0:30:45 The only restriction is, does it fit within the memory of the GPU?
0:30:47 But if it fits, it ships.
0:30:52 So, yes, this will run on GPUs that are out there in the market today.
0:30:53 Very cool.
0:30:53 Awesome.
0:30:55 Well, thank you very much.
0:30:57 This has been amazing, fascinating.
0:31:00 I love talking AI and nerd now, especially agents.
0:31:01 Agents is the hottest topic.
0:31:02 So, really appreciate you taking the time with me.
0:31:03 Absolutely.
0:31:05 I could also show you that in the future one day.
0:31:06 So, yeah, have a good time.
0:31:07 Thank you.
0:31:11 We’ll be right back to the next wave.
0:31:14 But first, I want to talk about another podcast I know you’re going to love.
0:31:18 It’s called Marketing Against the Grain, hosted by Kip Bodnar and Kieran Flanagan.
0:31:24 It’s brought to you by the HubSpot Podcast Network, the audio destination for business professionals.
0:31:30 If you want to know what’s happening now in marketing, what’s coming, and how you can lead the way, this is the podcast you want to check out.
0:31:34 They recently did a great episode where they show you how you can integrate AI into the workplace.
0:31:37 Listen to Marketing Against the Grain wherever you get your podcasts.
0:31:46 Hi, this is Bob Petty, Vice President and General Manager of Enterprise Platforms at NVIDIA.
0:31:59 Our first effort with AI developers, right, whether they’re enthusiasts, prosumers, or professionals, was basically using our RTX Pro cards or GeForce cards, running Linux.
0:32:05 And for many people to run Linux, we did Windows subsystem for Linux, WSL2.
0:32:21 Worked with Microsoft to eliminate a lot of what was intended for just an emulator and really turn it in, eliminate the latency blocks so that you could use it to truly develop AI and evaluate AI, not just from an accuracy standpoint, but also from a latency standpoint.
0:32:26 So historically, we’ve gone out with AI workstations or AI PCs for that.
0:32:26 Right.
0:32:30 If you’re running Intel or AMD CPU, put those in there.
0:32:39 The new RTX 6000 Pro Blackwell is a 96 gigabyte frame buffer, so you can run 70B models, fine-tune, do a lot of it.
0:32:47 A lot of the infrastructure that people are buying today in the cloud or through some of our server partners is Grace Blackwell.
0:32:50 It was Grace Hopper, now Grace Blackwell.
0:32:50 Right.
0:32:52 Grace being the ARM CPU, right?
0:32:52 Right.
0:32:58 And that’s kind of the same technology that all the big, big, like OpenAI, those types of companies are using.
0:32:58 Exactly.
0:33:03 And the beauty of that is the ARM CPU uses up so much less power.
0:33:15 So if the majority of the workload was in the GPU and the CPU is kind of a traffic manager, you wouldn’t be able to do what you needed to do for as little as power as possible so you can put more GPUs in there.
0:33:17 So hence, Grace Blackwell.
0:33:18 Right.
0:33:24 Well, you can certainly develop AI on Windows workstations with RTX Pro or GeForce.
0:33:24 That’s all great.
0:33:33 But if you need an ARM port of your software or getting familiar, that’s where Spark fits the gap.
0:33:36 And this is the same thing as the Project Digits that was announced at CES, right?
0:33:37 This is Project Digits, yeah.
0:33:39 We finally chose the name Spark.
0:33:41 We had people send in a lot of comments.
0:33:55 But as Jensen mentioned in the keynote, you know, what was a box that was like this several years ago, 20-core CPU, one petaflop is now in this little 5×5 by less than 2-inch box.
0:34:00 It’s got the C2C memory between the Grace processor and the Blackwell GPU.
0:34:02 It was like 256 gigabytes a second.
0:34:08 You won’t have that on a traditional workstation because you’re going to go over the PCI bus, right?
0:34:16 So high memory bandwidth between the CPU and the GPU, 128 gigs of memory available from us online.
0:34:20 But that’s really for the enthusiast who want the gorgeous bezel.
0:34:26 So what kind of things can I do with this now that I can’t do with my 5090 at home?
0:34:27 Good question.
0:34:32 So from a frame buffer size, there are so many more models that you can run on this.
0:34:34 You can do fine-tuning on 70B models.
0:34:37 You can put two of these together with this cable.
0:34:39 This is a ConnectX Ethernet.
0:34:42 Put two of these together, you can run a 400 billion parameter model.
0:34:44 You can’t do that on the 5090.
0:34:44 Right.
0:34:48 You can’t do the 70B on a 5090, right?
0:34:48 Right.
0:34:51 And so the size of the model is very dependent.
0:34:56 The other thing with a 5090, your memory bandwidth between your CPU and your GPU is throttled by the PCI bus.
0:34:57 Okay, right.
0:35:02 And this has, you know, cache-coherent high-speed memory bandwidth.
0:35:11 So it really enables you to test what your code might look like running on one of the data center providers, OEMs out there.
0:35:17 Because same C2C memory, cache-coherency, speed, you can do more than just say it works.
0:35:18 Right.
0:35:25 You can get it to the point where you can remove a lot of the bottlenecks, whether you’re doing a vision language model or, you know, multimodal model.
0:35:27 That’s the biggest benefit.
0:35:33 One is getting your code ready for what’s the predominant AI infrastructure out there.
0:35:33 Right.
0:35:40 But the other is testing in a way that simulates how it’s going to run, you know, when you run it on a node there.
0:35:48 And then the idea is you’re not wasting data center time or cloud time just debugging, right, or illuminating bottlenecks.
0:35:52 When you’re here, you deploy, and you scale from one GPU to end GPU.
0:36:00 So the main purpose of this was really to help spread the Grace Blackwell ecosystem.
0:36:00 Right.
0:36:06 They’re going as fast as we can make them, but they’re not necessarily accessible to enthusiasts.
0:36:07 Right.
0:36:16 Who want to use FP4 features of Blackwell, which you can do on 5090, but want to use it with some of the more popular models that have higher parameter sizes.
0:36:22 So EZBox, the one terabyte version of storage, 128 gigs, is $29.99.
0:36:25 The four terabyte version is $39.99.
0:36:29 You can reserve on NVIDIA.com.
0:36:33 Our initial go-to-market partners are Dell, HP, ASUS.
0:36:38 They’ve got their own branded boxes without the gold foil, and then we’ll expand that.
0:36:45 Yeah, I remember Jensen said something to the effect of, like, imagine it’s a cloud computer just sitting on your desk.
0:36:46 It’s not going to the cloud.
0:36:49 You know, you don’t have to worry about internet connections, anything like that.
0:36:54 It’s just a cloud computer sitting on your desk that can do all of the inference right there on the bigger models.
0:36:57 It’s an AI supercomputer on your desk, and there’s a bigger one that we’ll walk to in a second.
0:37:04 But the other thing about this is we’re not suggesting that everybody just replace their existing, you know, laptop or workstation.
0:37:08 You’ve got a GeForce laptop or RTX Pro laptop.
0:37:10 You plug this guy into it.
0:37:15 So you might do everything you need to on a 5090, run your games and everything.
0:37:20 You do an AI development or want to write an AI inferencing that helps you on your 5090.
0:37:21 Plug that into it.
0:37:34 And that’s why Jensen showed the MacBook and the Spark, because it’s really meant to be both a plug-in to Uber Assist, maybe less capable machines, whether they have a GPU in them or not.
0:37:41 And it’s not using all the processing on your computer, so you can be running AI models on this while playing Cyberpunk on your computer.
0:37:42 Exactly, yeah, exactly.
0:37:49 And, you know, just format factor-wise, cost-wise, we think it would be easy for people to do it as an add-on.
0:37:52 But certainly there’s, you know, there’s a great GPU in here.
0:37:54 We’re running games on this.
0:38:04 I wouldn’t want it as my GeForce laptop, but I’d probably want to connect it to my GeForce if I was, you know, doing AI tuning for game development or things like that.
0:38:09 So that’s DGX Spark, who has walked this way.
0:38:11 This is the RTX Pro line.
0:38:18 Our 6000 line is the one that’s, you know, somewhat akin to the 5090 on the GeForce side.
0:38:22 The reason we have this Pro line, the manufacturing of it is a very precise bomb.
0:38:25 It’s not built by many different AICs.
0:38:34 There are computing benefits on here that we perceive the gaming community doesn’t need, so some of the high-end compute performance here is going to be much better.
0:38:37 The AI inferencing performance is about the same.
0:38:38 The big difference is frame buffer.
0:38:39 Right.
0:38:42 And doesn’t this one have, like, 96 gigs of RAM?
0:38:44 Yeah, 96 gigs of RAM.
0:38:48 And if you want to get full power, we’ve got the 600-watt version.
0:38:53 We’ve got a 300-watt version that most desktops can take today.
0:38:58 And then you get the same technology in the server version that would go in a rack here.
0:39:07 So we used to call these, in the past, we’ve had A40s or L40s based on Ada Lovelace, the Ampere Lovelace.
0:39:17 We’re going to call that B40 based on Blackwell, but kind of aligned around RTX Pro, Workstation, Max-Q.
0:39:22 Max-Q is that optimal PowerPoint, and the server edition.
0:39:24 So, again, same infrastructure.
0:39:29 You can code and develop and then deploy on your RTX server and a rack in the data center, meant to save time.
0:39:32 So, these are our DJX stations.
0:39:39 Initial partners, Dell, HP, Asus, Lambda, and Supermicro.
0:39:42 We show these two here because these are their boxes.
0:39:43 This is what it will look like.
0:39:47 If you look in here, this is a GB300 board.
0:39:50 So, a much more powerful Grace processor.
0:39:54 And then an extremely powerful B300.
0:39:58 B300 is the same GPU in the latest DJX.
0:39:58 Okay.
0:40:03 So, 784 gigabytes of memory, again.
0:40:10 So, the benefit is you’re doing ARM development, a lot of memory, and very, very high-speed memory bandwidth like Spark.
0:40:18 So, you’re not just seeing that code works, you’re seeing how well it works before you chew up time on your data center rack.
0:40:22 Now, does this need like a separate CPU, like an Intel AMD kind of thing?
0:40:26 No, it’s on both the Spark and the station, we’re providing the Grace CPU.
0:40:27 Okay.
0:40:30 So, great CPU and the Blackwell GPU there.
0:40:31 Graphics out.
0:40:34 We don’t put a big graphics card in here.
0:40:35 So, this one has a 4,000.
0:40:37 That’s a small form factor.
0:40:41 The reason being, we want this to plug into a standard 15-amp wall outlet.
0:40:42 Right.
0:40:44 Might need to be a dedicated one.
0:40:47 Because, you know, it’s at 1,650 watts, I guess.
0:40:49 And we’re going to come pretty close to that.
0:40:52 Which is why the manufacturers are liquid cooling this.
0:40:55 And they started doing that in the gaming side.
0:40:57 So, some of the Alienware chassis, you’ve seen the liquid cool.
0:40:58 Right.
0:41:00 So, they’re well adept at that.
0:41:05 They will liquid cool this so we can stay, you know, thermally be good and not need to take
0:41:07 up a lot more power with a lot of fans.
0:41:14 So, again, if you’re an enthusiast, just getting started, even an enterprise where you know the
0:41:17 size of your models and what you want to get done, easily connect that.
0:41:24 If your full-time job is, like, prepping AI, developing AI to deploy to the data center,
0:41:27 you have your choice of logging it to the data center.
0:41:31 Maybe you get a virtual workstation delivered back to you and you do your job.
0:41:33 Or putting this at your desk.
0:41:33 Okay.
0:41:38 And this is, literally, it’s like one of those B300 nodes.
0:41:39 Right, right.
0:41:46 It is a server node in a desktop with graphics out, right there from a data privacy standpoint,
0:41:48 from an IP protection standpoint.
0:41:50 I’m not sending anything anywhere.
0:41:50 Right.
0:41:51 It’s right there.
0:41:56 And that’s more important than just privacy and IP protection.
0:42:00 It’s just the time and the cost of transport of data.
0:42:00 Right.
0:42:06 You might run a, you know, a 600 billion parameter model, but the data that you’re running it on,
0:42:11 whether let’s say it’s Cosmos, the BLM, all the videos that you’re going to be processing
0:42:19 and the amount of data, you’d want to just sit here versus upload all that data or even
0:42:22 if your own dedicated data center and run it there and then download that.
0:42:27 So you’ve got ingress and egress costs of data transport, all happening right here.
0:42:31 That guy will be available in the summertime frame.
0:42:31 Okay.
0:42:33 Reservable today.
0:42:34 This guy will be late summer.
0:42:35 Okay.
0:42:38 We have a founder’s edition for that because it’s cool.
0:42:38 Right.
0:42:41 Enthusiasts are going to want something on their desk, right?
0:42:43 This one is only available for the OEMs.
0:42:44 Okay.
0:42:46 The way for us to scale our enterprise businesses through the OEMs.
0:42:52 And so if you were to go to the Dell booth today or the HPI booth or Asus, you’ll see
0:42:53 their versions of these here.
0:42:59 You’ll see their versions of the Spark with the Dell blue and the NVIDIA green LED and the
0:42:59 HPE blue.
0:43:06 And that’s the way to expand the ecosystem for Grace Arm development, Grace Blackwell
0:43:11 development, expand the access to technology that normally is only available if you’ve got
0:43:13 the capital expense to put one of these racks in.
0:43:15 And now it’s at your desktop.
0:43:16 Very cool.
0:43:17 Yeah.
0:43:17 Amazing.
0:43:18 Well, thanks, Bob.
0:43:21 This has been really informative and I appreciate it.
0:43:21 Thank you.
0:43:21 Yeah, no problem.
0:43:22 Thank you.
0:43:41 Thank you.
0:00:08 I’m Matt Wolfe, and in today’s episode, we’re talking about AI agents.
0:00:15 I had the opportunity to go spend a few days with NVIDIA out at the NVIDIA GTC conference.
0:00:20 And in this episode, I’m going to deep dive with Amanda Saunders from NVIDIA.
0:00:26 She’s been at the heart of the AI agent revolution that’s happening right now while working with
0:00:30 NVIDIA, the company that’s enabling the AI revolution.
0:00:36 In this episode, we’ll unpack exactly what agentic AI is and how it’s reshaping industries
0:00:41 from healthcare and telecom to sports coaching, as well as why that’s just scratching the surface
0:00:42 of what’s possible.
0:00:48 You’ll also discover NVIDIA’s secret blueprint for easily building powerful AI agents, as
0:00:54 well as get into some of the fears that people have around AI agents, like is an AI agent going
0:00:56 to take your job and replace you?
0:00:57 Yeah, we’re going to go there.
0:01:01 And after the interview, I’m going to share some cool clips from NVIDIA’s GTC, where I
0:01:09 got Bob Petty, also from NVIDIA, to break down exactly what NVIDIA’s DGX Spark is and how
0:01:14 they’re going to be putting AI supercomputers in normal people’s homes.
0:01:18 You can have an AI supercomputer in your home by this time next year.
0:01:20 We’re going to get into that in today’s episode.
0:01:25 So without further ado, here’s my discussion with Amanda Saunders, followed by my tour of the
0:01:27 EGX Spark with Bob Petty.
0:01:28 Enjoy.
0:01:33 Marketing in 2025 is wild.
0:01:36 Savvy customers spot fake messaging instantly.
0:01:42 Privacy changes make ad targeting a nightmare, and everyone needs more content than ever.
0:01:46 That’s why you have to have HubSpot’s new Marketing Trends Report.
0:01:51 It doesn’t just show you what’s changing, it shows you exactly how to deal with it.
0:01:55 Everything’s backed by research, but focused on marketing plays you can use tomorrow.
0:01:58 Ready to turn marketing hurdles into results?
0:02:02 Go to click HubSpot.com slash marketing to download it for free.
0:02:07 Yeah, so my name’s Amanda Saunders.
0:02:11 I’m the director of enterprise AI software here at NVIDIA.
0:02:16 So you know a lot of people think about NVIDIA and all the amazing GPUs, CPUs, and DPUs that
0:02:18 we make that go into these powerful systems.
0:02:21 Well, actually, we have a lot of software that runs on that.
0:02:26 So me and my team, we focus on how do we tell that story and bring it to customers.
0:02:31 I’ve been at NVIDIA 10 years now, and I’ve covered everything from, you know, the graphic
0:02:34 side of our business into the data science.
0:02:37 And now, of course, the hottest topic, generative AI.
0:02:38 Very cool.
0:02:41 Let’s talk about agentic AI, because that’s not a hot topic.
0:02:45 It seems like in 2025, everybody’s saying this is the year of agentic AI.
0:02:48 How would you define agentic AI?
0:02:49 Absolutely.
0:02:52 So agents are helpful software tools.
0:02:56 They are digital employees that help us augment our work.
0:03:01 And what’s really important about agents and agentic AI is that it can perceive.
0:03:02 So it sees the world around it.
0:03:04 It sees the data that it has access to.
0:03:06 It takes that information.
0:03:08 It reasons about that information.
0:03:09 It thinks about it.
0:03:13 And then it can actually make actions based on that data.
0:03:16 So it can perceive, it can reason, and it can act.
0:03:20 And that’s what really sort of sets it apart from generative AI that we had in the past,
0:03:26 is being able to take those actions, whether that action is alerting a human or actually
0:03:30 actively using a tool and making something happen.
0:03:34 And so, yeah, I think they’re really exciting because they’re able to solve complex problems
0:03:37 that we’ve never been able to do with software before.
0:03:37 Very cool.
0:03:41 So what do you think makes agentic AI powerful right now?
0:03:42 Why not a year ago?
0:03:43 Why not two years ago?
0:03:46 Why is now the time everybody’s talking about it?
0:03:48 You know, I think this has been a journey that we’ve been on.
0:03:51 You know, it started with having to have the accelerated computing.
0:03:53 You had to have the computing power to do that.
0:03:56 And that’s a problem that NVIDIA has been working on for 30 years.
0:03:59 And then finally, we’ve got to the place where we have the computing systems.
0:04:01 We also then needed the software systems.
0:04:05 And so it started with the open models that became available.
0:04:09 I think LAMA was a huge advancement and step forward.
0:04:12 And more and more models have come out recently.
0:04:16 Reasoning models, in fact, are one of the big pieces that have sort of driven that
0:04:18 and sort of added to that.
0:04:22 And then just the full ecosystem of tools and things that are required.
0:04:28 I think AI is one of those really interesting fields where the more people use it, the more
0:04:31 they want to use more of it and they want to do more.
0:04:35 And so it’s really spurring this incredibly fast growth that we’re seeing, which is, you
0:04:36 know, breakneck speed.
0:04:42 But that’s what’s really driving us from, you know, the early onsets of generative AI,
0:04:46 where you’re just spitting out answers into these worlds now where they’re actually thinking
0:04:50 things through and making really smart actions, which is just sort of cool.
0:04:50 Yeah.
0:04:55 Or in my case, I got into AI and now I just make content about it because I’m obsessed with it.
0:04:56 It’s all I ever want to talk about.
0:05:00 Because yeah, once you get into it, you really, that’s all you want to do is you figure out
0:05:01 what it can do.
0:05:05 You learn where it’s, you know, boundaries are, and then you want to do more and more and
0:05:05 more.
0:05:10 And as it continues to, you know, get better, you can do more and more things with it.
0:05:10 So cool.
0:05:13 Well, can we walk through like a agentic workflow?
0:05:17 Like give some examples of like, this is the types of things an agent can do.
0:05:18 Yeah, absolutely.
0:05:23 I think, you know, some of the most basic things that we see from agents are really about
0:05:24 being able to talk to data.
0:05:28 I think that’s probably the first thing that we see most people try to do is they have their
0:05:34 own personal data, they connect an LLM to that data, and then they’re able to ask questions.
0:05:38 And I think, you know, more recently, they’re able to ask more and deeper questions.
0:05:40 Deep research has been a big topic.
0:05:42 It’s a big topic here at GTC.
0:05:48 And the more information you can get out of that data and the more quickly you can query that,
0:05:51 that’s a pretty standard agent that we would see out there.
0:05:54 Now, then they start to get a lot more complex.
0:05:59 I think there’s some really cool stuff going on in the telecom space where they’re actually
0:06:05 using agents to improve the network so they can actually predict when there might be network
0:06:12 outages coming and they can make recommendations to the human employees who, you know, maybe you
0:06:17 want to make these changes because maybe there’s a large show in town like GTC and there’s going
0:06:20 to be a lot of traffic on that network.
0:06:25 Here are the recommended changes so that you can actually, you know, still provide the service.
0:06:27 Is that kind of thing happening at GTC right now?
0:06:28 Are they using that?
0:06:29 I wish they would.
0:06:33 I think, I think the agents are just starting to be developed.
0:06:37 So maybe next year we’ll actually see that, but it’s definitely something that we’re going
0:06:37 to see more of.
0:06:38 Yeah.
0:06:38 But yeah.
0:06:41 So those I think are, you know, some varying levels of examples.
0:06:45 And then there’s obviously hundreds more in the healthcare space, you know, helping nurses
0:06:50 and doctors become more efficient because we know they need the help to, you know, everyday
0:06:51 people like you and I.
0:06:56 I mean, I use it for almost everything I do, whether it’s in my personal life or my
0:06:56 work life.
0:06:59 Are there any like agents that might surprise people?
0:07:04 Anything that’s like really sort of like interesting that, oh, I wouldn’t have thought people would
0:07:05 be using agents in that area.
0:07:07 It’s an interesting question.
0:07:12 I mean, the most interesting one for me on the agent side is really, as you started to
0:07:16 get into video and other types of data sources, like text, I think people have gotten the handle
0:07:21 on, but adding sort of video understanding and things coming through it.
0:07:25 Actually, one of the really cool use cases that we recently did was Jensen got to throw
0:07:27 out the first pitch at a baseball game.
0:07:28 I saw the clip of that.
0:07:36 And we used a video agent to be able to critique his performance, which I just think is really
0:07:36 cool.
0:07:41 So you can imagine for, you know, athletes, whether they’re amateurs or professionals, being
0:07:43 able to use that is really helpful.
0:07:43 Yeah.
0:07:44 Yeah.
0:07:46 I can see that like golfers and stuff like that.
0:07:47 Analyze my swing.
0:07:48 Tell me what I’m doing wrong.
0:07:48 Stuff like that.
0:07:53 And it’s probably too much for the AI on my swing, but you know, it still would be helpful
0:07:54 to know.
0:07:54 Yeah.
0:07:54 Yeah.
0:07:56 Well, let’s, let’s talk about the NVIDIA blueprints.
0:08:00 Because when I was at CES, that was a big topic during CES was the NVIDIA blueprints.
0:08:03 And to me, that’s sort of like the beginnings of agents, right?
0:08:06 But there’s sort of this pre-built agent.
0:08:07 You could probably explain it better than I can.
0:08:09 Well, exactly.
0:08:12 We tried to name them blueprints so that people would get the idea that these are reference
0:08:13 architectures.
0:08:18 And what they do is they take the different building blocks that NVIDIA is offering to
0:08:22 make these agents, and it helps you with a recipe on how to put them together.
0:08:25 So it starts out with our NIM microservices.
0:08:29 And so these are packaged, containerized, optimized models.
0:08:30 Right.
0:08:34 So all the leading open models that are out there in the market, we take them, we containerize
0:08:39 them, we add a standard API call to them so that people, you know, developers out there
0:08:40 can start building.
0:08:42 And we package those up.
0:08:43 So now you’ve got the model.
0:08:45 And now you need to connect that model.
0:08:46 Right.
0:08:46 It’s right.
0:08:47 A model and it’s only, only does so much.
0:08:48 So that’s…
0:08:54 So like a NIM would be like, you’ve got a video NIM that this one NIM can produce AI-generated
0:08:55 video.
0:08:57 This NIM could produce text-to-speech.
0:08:59 This NIM could do speech-to-text.
0:08:59 Exactly.
0:09:00 Right.
0:09:00 Okay.
0:09:00 Exactly.
0:09:05 So we actually have a hundred NIM that are now available that you can download and use.
0:09:10 And they do, there are 40 different sort of domains or modalities that these different
0:09:11 NIM can do.
0:09:12 So yeah.
0:09:14 So you can take a bunch of them, piece them together.
0:09:17 And so when we built those, everyone was like, this is great.
0:09:18 We need the models.
0:09:22 We need them to run really well on, you know, our NVIDIA GPUs.
0:09:24 But then they said, well, but how do we build them into the next thing?
0:09:26 And that’s where Blueprints came from.
0:09:26 Right.
0:09:31 So they are the recipe that allows you to take the NIM, start with this recipe, piece this
0:09:32 together, and you’ll get to an agent.
0:09:36 And then what’s even better is you can then customize them.
0:09:38 So you can, you know, add your own data sources.
0:09:40 You can add your own pieces in.
0:09:42 You can combine Blueprints together.
0:09:47 So maybe you started with a chatbot and you wanted to add a digital human on the front end.
0:09:49 We have a Blueprint for both of those.
0:09:52 You piece them together and all of a sudden you have a digital avatar.
0:09:57 Who can, you know, talk to you with a full, you know, face and expressions and natural language.
0:09:58 Probably even your voice if he wanted to.
0:09:59 Yeah.
0:09:59 Absolutely.
0:10:04 And you can, you can start with pieces from NVIDIA or you can start with pieces from the
0:10:04 ecosystem.
0:10:07 We try to make it really easy to piece it all together.
0:10:08 Super cool.
0:10:10 Let’s talk a little bit about like security.
0:10:15 I know when it comes to AI agents, they can be used for both good and bad.
0:10:19 What kind of things can we do to sort of protect and secure and make sure that people aren’t
0:10:23 going and using these agents for, you know, bad actor type stuff?
0:10:23 Absolutely.
0:10:30 I think, you know, what’s really cool about AI is where it raises concerns or security challenges
0:10:32 and it can actually also help answer them.
0:10:36 So there’s a Blueprint out there for container security that does a lot of those pieces.
0:10:43 But we also have other applications that can actually track activities and make alerts based
0:10:45 on detecting some anomaly, right?
0:10:49 So that’s tends to be how you recognize that something’s doing something it maybe wasn’t
0:10:50 supposed to.
0:10:54 And AIs are just really good at that because they can, you know, consume a lot of information.
0:11:00 So I think where, you know, AI potentially can raise some concerns, it’s also in some ways
0:11:03 the answer to addressing some of those concerns, which I think is really helpful.
0:11:03 Right.
0:11:09 So like the sort of like sci-fi movie scenario of like the AI rising up against us.
0:11:14 I mean, is this something that people should be concerned about or is this something that
0:11:16 you feel there’s pretty good guardrails in place for?
0:11:20 I think the builders of these apps need to make sure those guardrails are in place.
0:11:27 But I think, yes, in general, I think the tools exist and then it’s just about us as the application
0:11:29 builders being smart about how we deploy them.
0:11:30 Right.
0:11:33 You know, in general, I think agents are really powerful.
0:11:38 They’re incredible tools that help humans, but on their own, they’re not about to sort
0:11:39 of go off the rails.
0:11:41 It really takes a human to take it there.
0:11:43 So I think, again, it is a tool.
0:11:46 It is something that can, you know, work alongside you.
0:11:50 But I don’t see the rise of the machines quite yet.
0:11:55 So along similar lines, I think one of the fears a lot of people have is like, is an agent
0:11:56 going to take my job?
0:12:00 So, you know, what are your thoughts in that sort of realm?
0:12:06 I mean, for me personally, I see all of AI as like a tool, something that makes me a lot
0:12:09 more efficient and makes it so I can accomplish more in a day.
0:12:11 But I’m curious, like, what’s your take?
0:12:13 Do you think that, you know, people need to be concerned about that?
0:12:18 I think the best thing I’ve ever heard is an agent’s not going to take your job, but somebody
0:12:19 using an agent might.
0:12:21 And that’s always the thing that I think gets framed.
0:12:23 No, I think exactly that.
0:12:27 Like we as humans have the capacity to do so much more.
0:12:29 There’s just often so many hours in the day.
0:12:29 Right.
0:12:35 And so we don’t, you know, accomplish, you know, with agents, the work that we were trying
0:12:36 to do and then just stop.
0:12:36 Right.
0:12:39 No, it allows us to go and do more and try more.
0:12:42 And so I think, you know, for me, it’s not about replacing your job.
0:12:47 It’s about making you more effective at your job and being able to do more and, you know,
0:12:48 be more powerful.
0:12:53 I think, you know, healthcare for me is one of those industries that this is driven out there.
0:12:58 There’s actually a great partner of ours in the mental health space and they’re using
0:13:03 agents to free up therapist time from all the administrative work that they have to do.
0:13:05 We know their jobs are hard enough as it is.
0:13:11 If we can free that time up, it allows them to spend more time with their patients, actually
0:13:17 deeply understanding and working with them as opposed to worrying about calendars and taking
0:13:17 notes and scheduling.
0:13:22 So I think those are great examples where it’s, no, it’s something there that’s going
0:13:23 to help your job.
0:13:24 It’s not going to take your job.
0:13:24 Yeah.
0:13:28 I bumped into somebody on the street while I was walking yesterday and they had a very similar
0:13:29 concept for their company.
0:13:34 They work with therapists and they create chatbots specifically for the therapist so that their
0:13:38 customers can go and have the initial conversation with the chatbot.
0:13:40 But then the conversations get passed along to the therapist.
0:13:45 The therapist can kind of, you know, detect when some bigger issue that they need to address
0:13:47 comes up, but it’s not replacing the therapist.
0:13:50 It’s just, hey, I can now handle more patients, you know.
0:13:53 And give them more of my dedicated time.
0:13:53 Right.
0:13:57 And I think there’s something actually interesting as you start to see chatbots and particularly
0:14:03 those with digital avatars is humans will open up in some ways to one of these, you know,
0:14:07 chatbots or these avatars in ways that they may not necessarily feel comfortable opening up.
0:14:11 So I think it’s just giving us new avenues to do things that we were already doing anyway.
0:14:13 And I think that’s pretty cool.
0:14:13 Yeah.
0:14:17 This is sort of switching gears a little bit, but do you think there’s like a compute bottleneck
0:14:17 for agents?
0:14:21 Are we going to be able to scale agents and get to agents as fast as people would want
0:14:21 to?
0:14:25 There was a narrative, I feel like the narrative sort of faded a little bit, but there was
0:14:28 a narrative maybe six months ago that there’s a wall and AI is hitting a wall.
0:14:29 Do you think there’s a wall?
0:14:31 Do you think we’re going to hit some compute bottlenecks?
0:14:36 I think, you know, this is our life’s work at NVIDIA is to make sure that we have the compute
0:14:39 to, you know, drive the world’s agents.
0:14:44 And so I think, you know, a lot of the announcements we talked about today, both on the hardware and
0:14:49 on the software side are really focused on making these as efficient as possible.
0:14:55 And I think what’s really cool is as you’re looking at this AI space, when it first comes
0:14:58 out, it’s usually big, it’s boxy, it’s maybe not the most efficient.
0:15:03 And then over time, that efficiency comes in and allows us to do the next big leap.
0:15:06 And again, that starts out, you know, big and heavy.
0:15:09 And again, gets more and more efficient over time.
0:15:13 So I think I’m not seeing a bottleneck, but I do know it’s an area we need to continue
0:15:18 to drive because I do think the compute requirements are going to continue to grow.
0:15:18 Cool.
0:15:21 Well, so let’s extrapolate out a little bit.
0:15:24 So if you’re looking like five, 10 years down the road, what do you sort of envision with
0:15:25 AI agents?
0:15:26 Where do you think this is all headed?
0:15:29 And I know that’s hard because it’s sort of exponential technology.
0:15:31 It’s really hard for humans to grasp back credentials.
0:15:32 It’s incredibly hard.
0:15:36 I mean, I can barely keep six months out, you know, and new things keep popping up.
0:15:41 I mean, I think certainly the biggest things that we’re going to see are teams of agents.
0:15:44 And that’s going to be, you know, that we’re already seeing it today.
0:15:46 It’s starting now and it’s going to continue to grow.
0:15:52 And so, you know, the way that I see it is, you know, when we first got agents and we first
0:15:57 got, you know, chatbots and things, you could ask a question and it would respond to a question.
0:16:00 And then you could ask it to do a function and it could now do that function.
0:16:02 And then you can ask it to do a bigger job.
0:16:05 And these agents are just getting more and more sophisticated.
0:16:10 So five to 10 years out, you know, we may be able to give something, you know, as simple
0:16:14 as, you know, design me my, you know, retirement home.
0:16:18 And it will come back with all of the pieces involved in that without a human ever having
0:16:19 to do another prompt to follow it up.
0:16:21 And I think that could be pretty cool.
0:16:24 So that might even be just two or three years.
0:16:27 But yes, I certainly see that.
0:16:28 I think that’s where we’re going.
0:16:29 So you mentioned teams of agents.
0:16:30 That’s really fascinating to me.
0:16:35 Like, do you have any examples of like what sort of agents would team up to work together
0:16:36 and what sort of tasks will accomplish?
0:16:37 Yeah, absolutely.
0:16:38 So I work in marketing, right?
0:16:44 And so a lot of my job is about doing messaging, writing blogs, getting imagery, building
0:16:45 demo videos, things like that.
0:16:50 And today, you know, you start with one, you know, app that can help you write.
0:16:52 And then you go to another app that helps you build images.
0:16:55 And then you go to another app that helps you write the code for the website.
0:16:56 And then, you know, right.
0:16:58 And they’re all separate.
0:17:00 And each one of them returns to me.
0:17:03 And I, you know, move to the next step.
0:17:09 I think in the future, what we’re going to see is those agents will all be connected through
0:17:09 function calls.
0:17:14 And we actually, uh, at GTC announced a blueprint that’s going to help us do this.
0:17:15 It’s called IQ.
0:17:21 And so by connecting all these together, again, as a human, I’ll be able to start putting
0:17:23 my overall requests.
0:17:27 I need to build a new website for a new announcement that’s coming out and it will be able to do
0:17:29 all those functions together.
0:17:34 And I think what’s really cool about this is it’s going to allow us to design agents that
0:17:40 solve specific tasks, but by combining them together in that sort of that composable way,
0:17:43 they’ll be able to do bigger and bigger job functions.
0:17:43 Very cool.
0:17:49 So like what sort of bigger world plot problems do you see like AI and AI agents solving for
0:17:49 us?
0:17:54 I mean, I think networking is a really interesting one because anything that has that much data
0:18:01 that humans can’t solve, I think is, is amazing, you know, digital twins and simulation of those
0:18:06 types of environments, whether it’s, you know, the climate and weather, whether it’s, you know,
0:18:08 the businesses and things that we run.
0:18:13 I think all of these are areas where agents are just going to add to the ability to work with
0:18:14 these.
0:18:16 So, yeah, certainly I think those are big overall.
0:18:22 And then I think businesses, of course, it’s going to be about, you know, providing those
0:18:25 tools so that their employees are just that much more effective.
0:18:25 Yeah.
0:18:30 And that’s, that’s not a big world problem, but it’s a very common serious world problem.
0:18:30 Yeah.
0:18:34 One of the things that I think was really fascinating and I’ve heard, I’ve, I’ve made jokes about
0:18:38 how I feel like I’m on like the Jensen tour because I’ve actually seen his last like five
0:18:38 keynotes.
0:18:41 But one of the things that I’ve, I’ve seen him talk about that’s really fascinating to
0:18:47 me is the earth too, where it’s got the, they basically couldn’t map out the weather patterns
0:18:49 and figure out weather events a lot earlier.
0:18:54 So I’m really excited to see the sort of overlap of like the earth to concept and agents and solving
0:18:57 some of the more like bigger climate type issues as well.
0:18:58 A hundred percent.
0:19:03 I mean, I think one of the things when, when we introduced earth to is this idea that we,
0:19:08 we think a lot about these problems, we have conversations about them, but it’s really hard
0:19:13 as humans to sort of visualize what, you know, some of the changes that we make in government
0:19:16 and things like that are going to do in the next five to 10 years.
0:19:18 We’re very immediate creatures, right?
0:19:19 So that’s where we’re focused.
0:19:23 But I think, you know, imagine being able to go to earth too and through an agent say, Hey,
0:19:26 you know, if we made these changes, what would happen?
0:19:32 Or if we, you know, took these steps, how could that affect the world and what more can we
0:19:32 do?
0:19:37 And maybe there are things that we’re not even recognizing that the agent could recommend.
0:19:41 Because I think that’s, what’s really cool about these agents is they’re not humans.
0:19:43 They don’t think the way we do.
0:19:48 And so by giving them all that data and giving them an earth simulation, they might be able
0:19:49 to uncover things that we’ve never thought of.
0:19:50 Yeah.
0:19:50 Yeah.
0:19:51 I think that’s pretty cool.
0:19:51 Yeah.
0:19:52 I think that’s amazing.
0:19:57 What are some of the things in the AI world, you know, agents or otherwise that have you personally
0:19:58 really excited?
0:19:59 Like what sort of stuff do you use?
0:20:00 What do you play with?
0:20:02 Like, what’s your AI stack that you use?
0:20:05 I mean, I use as many as I can get my hands on.
0:20:08 We have a lot of them in NVIDIA.
0:20:12 I mean, one of my favorites, and I know Jensen talked about it a little bit on stage, is perplexity.
0:20:13 Yeah.
0:20:13 Oh, I love perplexity.
0:20:19 I love perplexity because I think it changes the dynamic of humans and how we interact.
0:20:25 Rather than searching for information and having to spend the thought process on that, it’s really
0:20:27 about who can ask the best questions.
0:20:33 And I think that’s a really powerful change and, you know, power dynamic that perplexity
0:20:34 is given to the users.
0:20:36 Now, if you can ask great questions, you can find great information.
0:20:38 So I love that tool.
0:20:44 We have AI agents in our company that help us with everything from, you know, our benefits
0:20:48 and understanding how to make the right decisions, you know, for each employee, which I think is
0:20:51 pretty cool, into, you know, how we do our jobs.
0:20:56 Whether that’s video creation, image creation, certainly content, which you have to write
0:20:57 a lot of.
0:20:58 So, yeah.
0:20:59 So all of those, I think, are great.
0:21:04 And then, you know, I take them into my personal life in terms of organizing and planning.
0:21:09 I’m very organized at work and I don’t always save a lot of that organization for my personal
0:21:10 life.
0:21:13 And so I can hand it off to, you know, any sorts of chatbots.
0:21:16 I think ChatTPT is excellent for this.
0:21:17 It’s just pretty cool.
0:21:18 And lately, deep research.
0:21:20 So I’ve been using a lot of that.
0:21:25 And again, we just introduced a blueprint that’s going to allow deep research on your
0:21:26 own personal things.
0:21:28 And so I can imagine that’s going to be pretty powerful.
0:21:33 So for people that want to sort of stay in the loop on AI that are curious, maybe they’re
0:21:33 worried about it.
0:21:37 Like, what sort of advice would you give them to sort of stay on top of things?
0:21:41 My best advice on AI is use it.
0:21:46 It sounds really obvious, but I think, one, I think it actually alleviates a lot of concerns
0:21:48 when you understand how the technology works.
0:21:54 And by using it, you can see what works today, where the limitations are, and how it kind
0:21:55 of functions.
0:22:02 And I think it lets people see it as that tool versus that, you know, something that might
0:22:02 be scary.
0:22:06 So that’s really the first piece of advice.
0:22:07 I think that’s really important.
0:22:12 And then I think it’s about, you know, trying to identify where many of those, you know,
0:22:17 concerns might be coming from, whether that’s data or security or things like that, and understand
0:22:19 how AI can also help with those.
0:22:23 And so that tends to be, you know, one of my best pieces of advice.
0:22:23 Cool.
0:22:24 Yeah.
0:22:26 Well, let’s talk about Llama 3 and Nemo Tron.
0:22:27 Yeah.
0:22:28 So Llama, Nemo Tron.
0:22:29 Llama, Nemo Tron.
0:22:30 Let’s talk about Llama, Nemo Tron.
0:22:31 Absolutely.
0:22:33 So Llama, Nemo Tron was a really cool announcement.
0:22:38 So NVIDIA, we work with all of the leading model builders that are out there, including Meta,
0:22:39 who released Llama.
0:22:46 And what’s really cool about Llama is there are, I think it’s 85,000 derivatives of this model.
0:22:51 And so, of course, NVIDIA, we’ve got a lot of really smart people who know how to optimize
0:22:53 and make models more efficient.
0:23:00 And when Reasoning came out and when DeepSeek introduced this reasoning wave into the open
0:23:06 source, we sort of said, well, how can we bring this and make it really efficient for,
0:23:08 you know, people who want to deploy this on NVIDIA?
0:23:15 And so by starting with the Llama model, which is an incredible model, we brought in our expertise
0:23:20 to train it so that a model that previously couldn’t do reasoning now could actually think
0:23:21 through problems.
0:23:26 And so we found, you know, the data set to go be able to train the model and how to do
0:23:27 this new task.
0:23:32 And we trained it and then we made the data source open, which I think is really cool.
0:23:36 So if others want to do their own training or if they want to train a different model or
0:23:38 anything like that, that data source is available.
0:23:44 But it’s just teaching a model a new skill is, I think, a really powerful thing to see because
0:23:47 it shows how quickly the space is evolving.
0:23:51 So is there anything actually happening like underneath Llama or is it just you’ve got
0:23:55 the Llama model, but now there’s a new layer on top of it that knows how to think?
0:23:57 Is there did any like new training happen to the model?
0:23:58 Yes.
0:24:00 So, yeah, the model was absolutely.
0:24:01 So we post train the model.
0:24:05 So this is what a lot of companies out there are doing today is they take a base model and
0:24:07 they actually train it with new data.
0:24:11 And in this case, sometimes you’re training it so it has more information on a particular
0:24:12 topic.
0:24:17 This is particularly popular when you’ve got domains and industries that have specific languages
0:24:20 that they speak, things like maybe the finance industry.
0:24:24 But in this case, we actually were training it on a skill.
0:24:27 And that skill is to think through problems.
0:24:31 And I think this is what’s really interesting about reasoning is the way a reasoning model
0:24:36 works is it starts by thinking through the question that you’re asking.
0:24:40 And it breaks down that question into multiple steps and multiple parts.
0:24:45 And then it actually goes through and comes up with answers for each of those parts.
0:24:51 And then it checks to say, OK, now that I’ve got those answers, does this actually, you know,
0:24:53 come back with the right question?
0:24:58 And it continues to do that until it gets to this highly accurate response.
0:25:05 And so not only are reasoning models really good for improving accuracy, which we all know
0:25:07 that when an AI model is useful, it’s when it’s accurate.
0:25:08 Right.
0:25:11 But it also allows us to solve problems we can never solve.
0:25:12 Right.
0:25:15 So a great example for me on this is I love puzzles.
0:25:20 I love all sorts of puzzles, but particularly Sudoku, because I hear it’s good for your brain.
0:25:27 Sudoku is a problem that humans can solve, but actually traditional LLMs couldn’t.
0:25:34 There are actually almost 80 different decisions that go into solving a Sudoku puzzle.
0:25:39 And each one of those affects the other decision, because obviously based on the rules that you
0:25:44 have to follow, reasoning models, and Lama Nibodron is a great example of this, can solve Sudoku.
0:25:45 Ooh, interesting.
0:25:48 Where a Lama model on its own couldn’t.
0:25:48 Wow.
0:25:50 And so again, that’s the type of thing.
0:25:54 It doesn’t sound like, you know, solving Sudoku puzzles is going to change the world.
0:25:59 But when you look at a Sudoku puzzle, there’s actually a lot of things that go on in the
0:26:01 world that are related to that.
0:26:02 Yeah.
0:26:03 I have a supply chain.
0:26:07 I’ve got to ship items and get them to different stores around the country.
0:26:12 I happen to know there’s a snowstorm coming in and I need to make sure that my trucks are
0:26:14 taking the most efficient routes.
0:26:18 All of those are steps that impact the other decisions that are being made.
0:26:23 And so it’s actually quite a complex problem that relates in some ways to Sudoku.
0:26:24 Yeah.
0:26:28 Well, it’s so interesting too, because you take a normal model and it can’t tell you
0:26:32 how many R’s are in strawberry, give it a thinking model, and it will actually count and then
0:26:33 double check.
0:26:34 Did I do that right?
0:26:34 And then answer.
0:26:36 And get the right answer.
0:26:36 Exactly.
0:26:40 It’s so fascinating because it seems like such a simple problem to a human brain.
0:26:44 But then that’s, I think, the thing is we look at models and we think of them like
0:26:46 we personify them as humans and they’re not.
0:26:47 Right.
0:26:49 And so, yeah, reasoning models are really cool for that.
0:26:53 And we’ve seen a lot of those great examples that come out of what these reasoning models
0:26:53 can do.
0:26:55 And it is.
0:26:57 It’s just, I think it’s really interesting to watch.
0:26:57 Yeah.
0:27:01 It’s so cool to actually see the thought process because you’ll actually see the models think
0:27:03 through something and then go, wait, that’s probably not right.
0:27:05 Let me think that through again.
0:27:08 And you actually see that text come out of it thinking through.
0:27:09 And that to me is fascinating.
0:27:10 It’s fascinating.
0:27:15 And it’s a great example of the compute story we were talking about, which is, you know, we do
0:27:17 need more compute so it can think.
0:27:19 The more it thinks, the more compute it requires.
0:27:24 So it’s this really interesting cycle that we’re watching these things go through.
0:27:28 One of my favorite things to do with reasoning models is to ask them to describe things to
0:27:29 me like a five-year-old.
0:27:29 Yes.
0:27:32 And it will come up with a description.
0:27:35 It will check if a five-year-old would understand that description.
0:27:36 It will then make changes.
0:27:39 And it’s really interesting to see what models think five-year-olds understand.
0:27:41 So at some point, I’ll have to go test it with a real question.
0:27:43 Yeah, read this.
0:27:44 Does this make sense to you?
0:27:45 Does this actually make sense?
0:27:46 Exactly, exactly.
0:27:47 Very cool.
0:27:50 Well, along the same line, real quick, let’s talk about hallucinations.
0:27:53 Do you see like a path to zero hallucinations?
0:27:55 Do we want to get rid of hallucinations completely?
0:27:56 Like, what are your thoughts on that?
0:28:01 I think, you know, there’s certainly things we can do to reduce hallucinations.
0:28:07 And some of those are, you know, as simple as putting a guardrail in place that says, if you’re
0:28:12 not 100% confident in the answer or 99% confident in the answer, don’t answer the question.
0:28:12 Right.
0:28:16 That sort of, it stops the model from doing things that it shouldn’t.
0:28:22 And Nemo guardrails, which is one of the products that NVIDIA offers, helps with building those.
0:28:26 But to your point about, do we want to stop hallucinations altogether?
0:28:31 Of course, we want it to give the right answers, but in being creative, we’re asking it to come
0:28:32 up with new things.
0:28:32 Right.
0:28:37 So, it’s, you have to be able to distinguish between a hallucination and a creative generative
0:28:37 response.
0:28:38 Right, right.
0:28:41 So, I think that’s where this balance plays.
0:28:47 So, there are absolutely steps that can be taken, and depending on how targeted and focused
0:28:51 you want the model to be, you can put more and more of those sort of guardrails or those
0:28:53 policies in place that will keep it from hallucinating.
0:28:54 Right.
0:28:57 Yeah, because I think in a lot of scenarios, hallucinations are a feature, not a bug, right?
0:29:01 If you wanted to write a short story for you, you wanted to hallucinate that short
0:29:01 story for you.
0:29:02 Exactly.
0:29:06 So, I think that’s where we have to understand what’s the hallucination versus what’s the model
0:29:07 doing what it’s supposed to do.
0:29:08 Right.
0:29:10 And again, and that’s where it becomes, what’s the use case?
0:29:12 What are you trying to have it do?
0:29:17 And really think those through, and then find the right tools from NVIDIA or others to be
0:29:19 able to actually, you know, go and do that.
0:29:20 Very cool.
0:29:24 Yeah, so if somebody wanted to get started with agentic AI, they want to start playing with
0:29:26 agents and testing the waters, what do they do?
0:29:27 What are our steps?
0:29:31 Well, so, from NVIDIA, we have something called build.nvidia.com.
0:29:33 We made the URL super easy.
0:29:36 If you’re trying to build something, we have a one-stop shop for you.
0:29:39 It’s got all the models on there so that you can test them out.
0:29:43 Whether it’s the new Llama Nemetron model with reasoning, you can actually turn reasoning on
0:29:43 and off.
0:29:45 It also has all the blueprints.
0:29:47 So, you can actually test them and experiment from them.
0:29:52 And then from there, there are also steps to go deploy them, test them out, and build them
0:29:53 yourself.
0:29:55 So, I think that’s a great starting point.
0:29:56 Very cool.
0:30:00 And for the more, like, technical people that maybe are trying to develop something, like,
0:30:02 is there a place they can go play with the APIs?
0:30:04 Like, what do we do there?
0:30:05 Build.nvidia.com.
0:30:05 Same place.
0:30:06 Same place.
0:30:11 It’s literally, whether you’re, you know, an enthusiast, whether you’re a developer, whether
0:30:16 you’re actually trying to, you know, build something to put in production, this is the one-stop
0:30:18 shop because everything’s on there.
0:30:20 You can take it as far as you want.
0:30:21 You can play around with the UI.
0:30:23 You can play around with the APIs.
0:30:28 You can actually download and deploy these models on any, you know, NVIDIA hardware.
0:30:30 It’s your one-stop shop for everything.
0:30:30 Yeah.
0:30:32 Actually, that brings me to another question.
0:30:35 Does this stuff work on older NVIDIA hardware?
0:30:40 If you have a, you know, a 3080 or a 4070, can I use this stuff on those as well?
0:30:41 Absolutely.
0:30:45 The only restriction is, does it fit within the memory of the GPU?
0:30:47 But if it fits, it ships.
0:30:52 So, yes, this will run on GPUs that are out there in the market today.
0:30:53 Very cool.
0:30:53 Awesome.
0:30:55 Well, thank you very much.
0:30:57 This has been amazing, fascinating.
0:31:00 I love talking AI and nerd now, especially agents.
0:31:01 Agents is the hottest topic.
0:31:02 So, really appreciate you taking the time with me.
0:31:03 Absolutely.
0:31:05 I could also show you that in the future one day.
0:31:06 So, yeah, have a good time.
0:31:07 Thank you.
0:31:11 We’ll be right back to the next wave.
0:31:14 But first, I want to talk about another podcast I know you’re going to love.
0:31:18 It’s called Marketing Against the Grain, hosted by Kip Bodnar and Kieran Flanagan.
0:31:24 It’s brought to you by the HubSpot Podcast Network, the audio destination for business professionals.
0:31:30 If you want to know what’s happening now in marketing, what’s coming, and how you can lead the way, this is the podcast you want to check out.
0:31:34 They recently did a great episode where they show you how you can integrate AI into the workplace.
0:31:37 Listen to Marketing Against the Grain wherever you get your podcasts.
0:31:46 Hi, this is Bob Petty, Vice President and General Manager of Enterprise Platforms at NVIDIA.
0:31:59 Our first effort with AI developers, right, whether they’re enthusiasts, prosumers, or professionals, was basically using our RTX Pro cards or GeForce cards, running Linux.
0:32:05 And for many people to run Linux, we did Windows subsystem for Linux, WSL2.
0:32:21 Worked with Microsoft to eliminate a lot of what was intended for just an emulator and really turn it in, eliminate the latency blocks so that you could use it to truly develop AI and evaluate AI, not just from an accuracy standpoint, but also from a latency standpoint.
0:32:26 So historically, we’ve gone out with AI workstations or AI PCs for that.
0:32:26 Right.
0:32:30 If you’re running Intel or AMD CPU, put those in there.
0:32:39 The new RTX 6000 Pro Blackwell is a 96 gigabyte frame buffer, so you can run 70B models, fine-tune, do a lot of it.
0:32:47 A lot of the infrastructure that people are buying today in the cloud or through some of our server partners is Grace Blackwell.
0:32:50 It was Grace Hopper, now Grace Blackwell.
0:32:50 Right.
0:32:52 Grace being the ARM CPU, right?
0:32:52 Right.
0:32:58 And that’s kind of the same technology that all the big, big, like OpenAI, those types of companies are using.
0:32:58 Exactly.
0:33:03 And the beauty of that is the ARM CPU uses up so much less power.
0:33:15 So if the majority of the workload was in the GPU and the CPU is kind of a traffic manager, you wouldn’t be able to do what you needed to do for as little as power as possible so you can put more GPUs in there.
0:33:17 So hence, Grace Blackwell.
0:33:18 Right.
0:33:24 Well, you can certainly develop AI on Windows workstations with RTX Pro or GeForce.
0:33:24 That’s all great.
0:33:33 But if you need an ARM port of your software or getting familiar, that’s where Spark fits the gap.
0:33:36 And this is the same thing as the Project Digits that was announced at CES, right?
0:33:37 This is Project Digits, yeah.
0:33:39 We finally chose the name Spark.
0:33:41 We had people send in a lot of comments.
0:33:55 But as Jensen mentioned in the keynote, you know, what was a box that was like this several years ago, 20-core CPU, one petaflop is now in this little 5×5 by less than 2-inch box.
0:34:00 It’s got the C2C memory between the Grace processor and the Blackwell GPU.
0:34:02 It was like 256 gigabytes a second.
0:34:08 You won’t have that on a traditional workstation because you’re going to go over the PCI bus, right?
0:34:16 So high memory bandwidth between the CPU and the GPU, 128 gigs of memory available from us online.
0:34:20 But that’s really for the enthusiast who want the gorgeous bezel.
0:34:26 So what kind of things can I do with this now that I can’t do with my 5090 at home?
0:34:27 Good question.
0:34:32 So from a frame buffer size, there are so many more models that you can run on this.
0:34:34 You can do fine-tuning on 70B models.
0:34:37 You can put two of these together with this cable.
0:34:39 This is a ConnectX Ethernet.
0:34:42 Put two of these together, you can run a 400 billion parameter model.
0:34:44 You can’t do that on the 5090.
0:34:44 Right.
0:34:48 You can’t do the 70B on a 5090, right?
0:34:48 Right.
0:34:51 And so the size of the model is very dependent.
0:34:56 The other thing with a 5090, your memory bandwidth between your CPU and your GPU is throttled by the PCI bus.
0:34:57 Okay, right.
0:35:02 And this has, you know, cache-coherent high-speed memory bandwidth.
0:35:11 So it really enables you to test what your code might look like running on one of the data center providers, OEMs out there.
0:35:17 Because same C2C memory, cache-coherency, speed, you can do more than just say it works.
0:35:18 Right.
0:35:25 You can get it to the point where you can remove a lot of the bottlenecks, whether you’re doing a vision language model or, you know, multimodal model.
0:35:27 That’s the biggest benefit.
0:35:33 One is getting your code ready for what’s the predominant AI infrastructure out there.
0:35:33 Right.
0:35:40 But the other is testing in a way that simulates how it’s going to run, you know, when you run it on a node there.
0:35:48 And then the idea is you’re not wasting data center time or cloud time just debugging, right, or illuminating bottlenecks.
0:35:52 When you’re here, you deploy, and you scale from one GPU to end GPU.
0:36:00 So the main purpose of this was really to help spread the Grace Blackwell ecosystem.
0:36:00 Right.
0:36:06 They’re going as fast as we can make them, but they’re not necessarily accessible to enthusiasts.
0:36:07 Right.
0:36:16 Who want to use FP4 features of Blackwell, which you can do on 5090, but want to use it with some of the more popular models that have higher parameter sizes.
0:36:22 So EZBox, the one terabyte version of storage, 128 gigs, is $29.99.
0:36:25 The four terabyte version is $39.99.
0:36:29 You can reserve on NVIDIA.com.
0:36:33 Our initial go-to-market partners are Dell, HP, ASUS.
0:36:38 They’ve got their own branded boxes without the gold foil, and then we’ll expand that.
0:36:45 Yeah, I remember Jensen said something to the effect of, like, imagine it’s a cloud computer just sitting on your desk.
0:36:46 It’s not going to the cloud.
0:36:49 You know, you don’t have to worry about internet connections, anything like that.
0:36:54 It’s just a cloud computer sitting on your desk that can do all of the inference right there on the bigger models.
0:36:57 It’s an AI supercomputer on your desk, and there’s a bigger one that we’ll walk to in a second.
0:37:04 But the other thing about this is we’re not suggesting that everybody just replace their existing, you know, laptop or workstation.
0:37:08 You’ve got a GeForce laptop or RTX Pro laptop.
0:37:10 You plug this guy into it.
0:37:15 So you might do everything you need to on a 5090, run your games and everything.
0:37:20 You do an AI development or want to write an AI inferencing that helps you on your 5090.
0:37:21 Plug that into it.
0:37:34 And that’s why Jensen showed the MacBook and the Spark, because it’s really meant to be both a plug-in to Uber Assist, maybe less capable machines, whether they have a GPU in them or not.
0:37:41 And it’s not using all the processing on your computer, so you can be running AI models on this while playing Cyberpunk on your computer.
0:37:42 Exactly, yeah, exactly.
0:37:49 And, you know, just format factor-wise, cost-wise, we think it would be easy for people to do it as an add-on.
0:37:52 But certainly there’s, you know, there’s a great GPU in here.
0:37:54 We’re running games on this.
0:38:04 I wouldn’t want it as my GeForce laptop, but I’d probably want to connect it to my GeForce if I was, you know, doing AI tuning for game development or things like that.
0:38:09 So that’s DGX Spark, who has walked this way.
0:38:11 This is the RTX Pro line.
0:38:18 Our 6000 line is the one that’s, you know, somewhat akin to the 5090 on the GeForce side.
0:38:22 The reason we have this Pro line, the manufacturing of it is a very precise bomb.
0:38:25 It’s not built by many different AICs.
0:38:34 There are computing benefits on here that we perceive the gaming community doesn’t need, so some of the high-end compute performance here is going to be much better.
0:38:37 The AI inferencing performance is about the same.
0:38:38 The big difference is frame buffer.
0:38:39 Right.
0:38:42 And doesn’t this one have, like, 96 gigs of RAM?
0:38:44 Yeah, 96 gigs of RAM.
0:38:48 And if you want to get full power, we’ve got the 600-watt version.
0:38:53 We’ve got a 300-watt version that most desktops can take today.
0:38:58 And then you get the same technology in the server version that would go in a rack here.
0:39:07 So we used to call these, in the past, we’ve had A40s or L40s based on Ada Lovelace, the Ampere Lovelace.
0:39:17 We’re going to call that B40 based on Blackwell, but kind of aligned around RTX Pro, Workstation, Max-Q.
0:39:22 Max-Q is that optimal PowerPoint, and the server edition.
0:39:24 So, again, same infrastructure.
0:39:29 You can code and develop and then deploy on your RTX server and a rack in the data center, meant to save time.
0:39:32 So, these are our DJX stations.
0:39:39 Initial partners, Dell, HP, Asus, Lambda, and Supermicro.
0:39:42 We show these two here because these are their boxes.
0:39:43 This is what it will look like.
0:39:47 If you look in here, this is a GB300 board.
0:39:50 So, a much more powerful Grace processor.
0:39:54 And then an extremely powerful B300.
0:39:58 B300 is the same GPU in the latest DJX.
0:39:58 Okay.
0:40:03 So, 784 gigabytes of memory, again.
0:40:10 So, the benefit is you’re doing ARM development, a lot of memory, and very, very high-speed memory bandwidth like Spark.
0:40:18 So, you’re not just seeing that code works, you’re seeing how well it works before you chew up time on your data center rack.
0:40:22 Now, does this need like a separate CPU, like an Intel AMD kind of thing?
0:40:26 No, it’s on both the Spark and the station, we’re providing the Grace CPU.
0:40:27 Okay.
0:40:30 So, great CPU and the Blackwell GPU there.
0:40:31 Graphics out.
0:40:34 We don’t put a big graphics card in here.
0:40:35 So, this one has a 4,000.
0:40:37 That’s a small form factor.
0:40:41 The reason being, we want this to plug into a standard 15-amp wall outlet.
0:40:42 Right.
0:40:44 Might need to be a dedicated one.
0:40:47 Because, you know, it’s at 1,650 watts, I guess.
0:40:49 And we’re going to come pretty close to that.
0:40:52 Which is why the manufacturers are liquid cooling this.
0:40:55 And they started doing that in the gaming side.
0:40:57 So, some of the Alienware chassis, you’ve seen the liquid cool.
0:40:58 Right.
0:41:00 So, they’re well adept at that.
0:41:05 They will liquid cool this so we can stay, you know, thermally be good and not need to take
0:41:07 up a lot more power with a lot of fans.
0:41:14 So, again, if you’re an enthusiast, just getting started, even an enterprise where you know the
0:41:17 size of your models and what you want to get done, easily connect that.
0:41:24 If your full-time job is, like, prepping AI, developing AI to deploy to the data center,
0:41:27 you have your choice of logging it to the data center.
0:41:31 Maybe you get a virtual workstation delivered back to you and you do your job.
0:41:33 Or putting this at your desk.
0:41:33 Okay.
0:41:38 And this is, literally, it’s like one of those B300 nodes.
0:41:39 Right, right.
0:41:46 It is a server node in a desktop with graphics out, right there from a data privacy standpoint,
0:41:48 from an IP protection standpoint.
0:41:50 I’m not sending anything anywhere.
0:41:50 Right.
0:41:51 It’s right there.
0:41:56 And that’s more important than just privacy and IP protection.
0:42:00 It’s just the time and the cost of transport of data.
0:42:00 Right.
0:42:06 You might run a, you know, a 600 billion parameter model, but the data that you’re running it on,
0:42:11 whether let’s say it’s Cosmos, the BLM, all the videos that you’re going to be processing
0:42:19 and the amount of data, you’d want to just sit here versus upload all that data or even
0:42:22 if your own dedicated data center and run it there and then download that.
0:42:27 So you’ve got ingress and egress costs of data transport, all happening right here.
0:42:31 That guy will be available in the summertime frame.
0:42:31 Okay.
0:42:33 Reservable today.
0:42:34 This guy will be late summer.
0:42:35 Okay.
0:42:38 We have a founder’s edition for that because it’s cool.
0:42:38 Right.
0:42:41 Enthusiasts are going to want something on their desk, right?
0:42:43 This one is only available for the OEMs.
0:42:44 Okay.
0:42:46 The way for us to scale our enterprise businesses through the OEMs.
0:42:52 And so if you were to go to the Dell booth today or the HPI booth or Asus, you’ll see
0:42:53 their versions of these here.
0:42:59 You’ll see their versions of the Spark with the Dell blue and the NVIDIA green LED and the
0:42:59 HPE blue.
0:43:06 And that’s the way to expand the ecosystem for Grace Arm development, Grace Blackwell
0:43:11 development, expand the access to technology that normally is only available if you’ve got
0:43:13 the capital expense to put one of these racks in.
0:43:15 And now it’s at your desktop.
0:43:16 Very cool.
0:43:17 Yeah.
0:43:17 Amazing.
0:43:18 Well, thanks, Bob.
0:43:21 This has been really informative and I appreciate it.
0:43:21 Thank you.
0:43:21 Yeah, no problem.
0:43:22 Thank you.
0:43:41 Thank you.
Episode 53: What role will AI agents play in addressing global challenges? Join Matt Wolfe (https://x.com/mreflow) Amanda Saunders (https://x.com/amandamsaunders), Director of Enterprise Generative AI Product Marketing at Nvidia, then Bob Pette (https://x.com/RobertPette) Vice President and General Manager of Enterprise Platforms at Nvidia, as they delve into the transformative potential of agentic AI at the Nvidia GTC Conference.
This episode explores the concept of AI agents as digital employees that perceive, reason, and act, reshaping industries like healthcare and telecom. Discover Nvidia’s approach to building powerful AI agents and the measures in place to ensure their secure and productive deployment. From optimizing workflows with agentic AI blueprints to fascinating agent applications in sports coaching, the discussion unpacks AI’s promising future.
Check out The Next Wave YouTube Channel if you want to see Matt and Nathan on screen: https://lnk.to/thenextwavepd
—
Show Notes:
- (00:00) Exploring Nvidia’s AI Revolution
- (03:29) AI’s Breakneck Growth Spurs Innovation
- (06:29) Video Agents Enhancing Athletic Performance
- (09:46) AI: Problem Solver and Concern Raiser
- (14:54) Rise of Sophisticated AI Agents
- (18:21) Earth-2: Visualizing Future Changes
- (21:53) Nvidia Optimizes Llama for Reasoning
- (23:50) Reasoning Models Enhance Problem Solving
- (27:20) Balancing AI Creativity and Accuracy
- (30:31) Nvidia’s AI Development in Windows
- (34:16) AI Development Acceleration Benefits
- (37:32) High-Power Servers & Workstations Overview
- (39:37) Liquid Cooling in AI Workstations
—
Mentions:
- Get the free AI Agent Playbook: https://clickhubspot.com/ovw
- Amanda Saunders: https://www.linkedin.com/in/amandamsaunders/
- Bob Pette: https://www.linkedin.com/in/bobpette/
- Nvidia: https://www.nvidia.com/en-us/
- Nvidia GTC Conference: https://www.nvidia.com/gtc/
- Earth-2: https://www.nvidia.com/en-us/high-performance-computing/earth-2/
—
Check Out Matt’s Stuff:
• Future Tools – https://futuretools.beehiiv.com/
• Blog – https://www.mattwolfe.com/
• YouTube- https://www.youtube.com/@mreflow
—
Check Out Nathan’s Stuff:
- Newsletter: https://news.lore.com/
- Blog – https://lore.com/
The Next Wave is a HubSpot Original Podcast // Brought to you by Hubspot Media // Production by Darren Clarke // Editing by Ezra Bakker Trupiano