AI Agents Helping Robots, Amazon Acquires Robotics Startup, White House Unveils AI Legislation

0:13Laura Mandaro:Welcome everyone to The Information's TI TV. My name is Akash Pasricha. It is Friday, March 20th. We are kicking off the show today with a bit of a recap on what the big robotics insights were that came out of NVIDIA's GTC conference this week. We'll then pivot to some exclusive reporting the information published about a robotics acquisition that Amazon has made. This week on the Editor's Cut, we're talking about the White House's AI policy framework released today, and also the big headlines from NVIDIA, Meta, Amazon, and also AI startups this week. And we'll wrap the show with our big read.

0:49Laura Mandaro:The Information's Microsoft reporter wrote an excellent profile on Asha Sharma, the company's new CEO of its$24 billion gaming business. It's going to be a fun show, so let's get right on into it. It was a big week of news this week coming out of NVIDIA's GTC. And while chips and agents and autonomous vehicles have gotten a lot of attention, robotics is one area that hasn't been talked about as much. Our AI reporter, Rocket Drew, had a conversation with NVIDIA's robotics chief at the event, and I want to bring him on to debrief all of that. Rocket, welcome back to the show. It's great to have you here.

1:27Thanks, Akash. Great to be here. Was it your first GTC this week? It was. It was. And it was quite a scene. Have you ever been? No, never. I recommend it, just for the anthropology of it all. Yeah, not the anthropic of it all. Not this time. um i i saw a funny uh it's friday i'm just gonna tell the story there was a funny tweet somebody was standing in front of anthropology the the the retail store and said going to try out claude and

1:56Laura Mandaro:i thought it was a very funny joke okay let's talk robotics so you had a conversation with nvidia's robotics chief and it's kind of interesting i haven't heard too much about the robotics insights that have come out of GTC. What were the big questions that you went into this conversation with? And then we'll talk a little bit about what you learned. Yeah, that's right. I mean, there hasn't been a lot of robotics news out of GTC this year. And in fact, a couple of roboticists I spoke to felt like robotics was a little bit overshadowed by some other action at the conference. Chips, of course, but you would expect that, given that's NVIDIA's whole business.

2:34but also the open claw craze, like digital agents that work on your computer stole the spotlight from robotics a little bit. So I think some roboticists had that impression coming out of Jensen's keynote that he was playing down robotics a little bit and had the same impression walking around the expo floor and seeing the actual robotic demonstrations that were on display. Now, it's only fair. I mean, CES was a couple months ago, and that has become such a robotics show. So I think for some people, that set the expectation. They wanted GTC to be just as robotics heavy as CES. And then it sort of fell short of the mark.

3:10But there were a few small announcements out of the conference. So one of those is that NVIDIA produces physical AI models. One of those is Cosmos, which is a world model that generates, say, videos, sort of realistic, physically realistic videos mostly. And then they have Groot, which is a physical AI model for powering humanoid robots. So at this conference, they released customized versions of those, Cosmos H and Groot H, which are customized to work specifically on healthcare data, for example, to power surgical robots. So they've collected a bunch of surgery data, they've customized the models on that surgery data, and now they've open sourced them.

3:50So it's a little bit niche, but those were some of the bigger robotics announcements out of the conference. okay so they make these announcements and the conversation that you have with the robotics chief is this a press conference this is a one-on-one interview yeah this was a one-on-one they they smuggled me through a hotel and through a back hallway and into this room where we waited silently in a corridor and i got whisked into this room and there was this conference room in a hotel kind of on the outskirts of the conference so we had a conversation there for about half an hour. And one of my first questions going in was, you know, have agents kind of stolen the show?

4:28And one of the things that I came to realize over the course of the conversation is that, well, there's a way to look at it where AI agents will also be a boon to robotics. There are ways that AI agents can speed up progress in robotics. So roboticists should also feel like they can benefit to some extent from the progress that's happening in agents and all of the excitement around OpenClaw. How is that? Explain that. I mean, I'm just wrapping my head around agents. Now the agents are going to control robots too? That was exactly my question. Well, what tipped me off to begin with that this was, you know, sort of a possibility was that walking through the expo hall, they had a demonstration where they showed off, here's how OpenClaw, though in this case it was NemoClaw, which is NVIDIA's version.

5:13Here's how it can help you file your taxes. You know, it had access to some tax documents, you asked it questions. And the way they spiced up this demonstration is they had a reachy mini robot, you know, they're about like yay tall, maybe a little under a foot tall, 3d printed looking robots from Hugging Face with these cute little antennas. And the way they gave you this demonstration of the open claw agent running on your computer was that you talk about your taxes to this robot. It's like, they didn't have to do that. But it was a way that the agent sort of took on a physical presence. So I was was like, okay, like maybe there's some kind of connection going on here.

5:48So asking about that, I think the two main connections that came to light are one, you could use these kinds of computer using agents or coding agents like Claude Code and Codex to develop the robotic models, to, you know, gather the training data, to run simulations, to train the models, to test the models. Like the coding agents can take on some of the role of what a researcher or an engineer in robotics does today to develop these models. So that's one thing, and maybe speed up that process a little bit. And the second is that in the future, you could imagine one agent, say, running a fleet of robots in a warehouse.

6:25And when a task comes in, like, oh man, we got all these orders, we got to get them from all these shelves into all these boxes, an agent could coordinate some of that work. It could say, okay, you humanoid robot, you're going to take this job. And you robot dog, you're going to walk around the perimeter and make sure everyone's... And you could coordinate the robots and you can make sure they're sharing information with each other. And that's sort of a future direction that agents could go as well. So, I mean, these are all very grand ambitions that I think are very exciting. The idea that, hey, we right now have these smart speakers and that's sort of the closest we have to talking to a robot in real time.

7:01Now we're saying, hey, you have a robot that can do things for you. Maybe I can talk to the Roomba and tell it to, I don't know, do like a Strava-like formation on the floor and actually be productive. Fine,

7:12Laura Mandaro:I get it. Did you ask her any questions about where the technology is today? Who's actually paying for these tools? Whether or not NVIDIA's chips are good enough to support these workloads right now? Yeah, absolutely. I think everyone sort of acknowledges that it's early, that we're in the early innings of this game, but NVIDIA is happy to start supporting the ecosystem even though it's early. That's sort of the way they would position it. So yeah, again, I was talking to Deepu Tala, who is NVIDIA's vice president of robotics and edge AI. He's sort of NVIDIA's most senior robotics head. And I asked him the same question, especially for Cosmos, this world model that can generate videos.

7:55It's a really exciting direction. We've been hearing a lot of enthusiasm for world models for a while, but often we also hear that they're just not quite there yet. They're not realistic enough in the videos that they generate to, say, train a robot that could then transfer into working in the real world. So I asked him the same question, when is it going to get there? And his answer is it really depends on the customer. Cosmos can generate different kinds of videos, it can generate different kinds of data, and depending on the customer's needs, it might be there already, especially if they just want to tinker around with it for research purposes, or it might take a little while to get there.

8:30But I think NVIDIA, has been making these kinds of long-term plays for a while, for example, in self-driving, which has been an interest for the company for a decade, even though they're just now declaring that self-driving has hit a sort of chat GPT moment. So I think that inflection point could still be delayed for a number of years, but they want to make some early progress in that direction. I wonder, as you think about your week at GTC more broadly, what were the questions that you didn't quite get as many answers to or the further questions that you walked away with saying, hey, I'm seeing a lot of potential here.

9:06I see the inference chips. I see NemoClaw.

9:10Laura Mandaro:Maybe I'm unclear on customer adoption. Maybe I'm unclear on business model. I mean, what sort of questions did you walk away with wanting more information on? Yeah, absolutely. I think, well, one thing was that there was a lot of enthusiasm for Nemo Claw, which is NVIDIA's version of OpenClaw. But one of the selling points is that it should be relatively easy to set up on your computer relative to OpenClaw. Some of the pain of setting up the software should just be taken care of for you. And I haven't seen that myself yet. So I walked away wondering how easy is it really to set up this software, especially when during the conference, Anthropic announced Dispatch, which is a similar feature for Claude Cowork, where you can sort of start jobs from your phone and it'll work on them on your computer and so on, which is one of the key selling points of OpenClaw.

10:02So I went to the tent, the Build-A-Claw tent, I guess, sort of Build-A-Bear that NVIDIA was hosting. And they had a lot of NVIDIA employees there helping conference goers set up NemoClaw on their devices and show conference goers how they could use NemoClaw. And a couple of my takeaways there were, one, people feel kind of hesitant to set it up themselves. They want to know what it can do for them first. They need someone to sit down and hold their hand and walk them through. These are some different ways you could use it. And then two, in setting it up, they seem to want some support from NVIDIA.

10:34It's not the most straightforward thing. And then three, to set it up, many people were either bringing their own specialized computers or buying them right on the spot from NVIDIA. So that's another question on my mind of how much people will really adopt NemoClaw or these personal agents in general right now if it requires getting a Mac Mini or buying an NVIDIA DGX Spark. Now, some conference goers were happy to buy those computers. I heard that the gift store sold out of these DGX Spark computers on the first day, and then NVIDIA employees stayed up all night installing NemoClaw on some other computers.

11:13And then the next day, before the end of the day, someone told me they had sold out of 31 of them. And these things cost thousands of dollars. So people are interested in being able to use it, but I'm curious how hard it will be for them to set up in practice. Great. Well, Rocket, I want to thank you for coming on. That is Rocket Drew, our AI and robotics reporter here at The Information. Thanks, Akash. The Information published exclusive reporting that Amazon has acquired autonomous robotics startup River. Joining me now to discuss that story is Catherine Perloff, one of the reporters behind that piece.

11:46Catherine, welcome to the show. It's great to have you back. Hey. Okay, what is River, the robotics company? So River, they make these robots that they kind of look like, I don't know, like kind of like, they look like little four-legged animals, I guess. They're kind of like a long box on top of legs with wheels that helps them navigate stairs and pathways. They're designed for last mile delivery. So getting the package from the vehicle it's being delivered in to the door. Um, and, uh, they've previously had partnerships with Just Eat Takeaway, um, which is a food delivery company and Veeho, which is also in the delivery logistics space.

12:36Um, but now they're being acquired by Amazon. Okay. And the idea here is, I mean, they're autonomous. So like, do, do they get released in fleets and then they pick up the orders and then drop them off? Is it, is that the idea? So they can't drive cars.

12:53Laura Mandaro:No, I know that. I mean, I understand that. I mean, but they just get dropped off in fleets, and then they just go crazy. Yeah, I don't know if it'd be a fleet. It might just be like, I guess that's a question. I'm not exactly sure how they'd be deployed. It might be like one in a car, maybe two in a car. But the idea is like the driver could go drop off a package to one house, and then the river robot could go to a different house at the same time, and then you could deliver more packages quicker or something like that. Okay. Tell me about Amazon's robotics strategy at large. How does this fit into that?

13:32Has it made other acquisitions in the past that this is planning to be added to? Yeah. So Amazon has a long history in robotics. it has actually previously invested in the company via this thing called like their industrial fund where they invest in a lot of robotics companies. I think what makes the river kind of acquisition a little bit unique is a lot of the companies they've acquired and invested and have been more on the logistics side of things. And so, or sorry, this is, I guess, in the warehouses, I should say. So like, you know, helping, you know, move boxes, pack boxes, et cetera, and a little bit less in the delivery side of the process.

14:12You know, I guess they did invest in Rivian, which is an electronic electric vehicle company. They make electric trucks. Then Rivian is, you know, developing self-driving capabilities. Uh, Amazon also, uh, is behind Zoox, which is a, um, uh, self-driving robo-taxi, but that's really more for, like, people taking rides than it is for packages and delivery. So this is sort of like a delivery-focused robotics acquisition. And I think their previous efforts have more been about robots in the warehouses, robots in the factories, and also robots, like the AI behind the robots, the software and the models behind the robots.

14:58And how does the company ultimately talk about its ambition with robots and autonomous vehicles even? because the delivery personnel that Amazon has, I mean, it's a huge part of their workforce, right? I mean, they are known to have people in every corner of the world delivering things.

15:18Laura Mandaro:Are they openly saying we want to replace them to lower costs or what do they say about that? Yeah, certainly not openly. I feel like they're very sensitive about this. The statement they gave to us about the acquisition was kind of like the robots will work with our delivery drivers. And, you know, I think it's always a kind of an interesting question with robotics, like can it make things safer for the people? But, and I think also at this stage, to develop this robotics and also to develop the AI behind the robotics can be really expensive. So, you know, in some ways it can be a cost center, but do I think they look down the line and say, can we employ less people?

16:04I mean, also with their delivery drivers, they don't even directly employ them. It's through this kind of controversial contracting model. So I think they're probably safe to say they're looking to cut costs. But I would, you know, maybe not people totally out of the loop yet. And last question for you. I mean, what is your assessment of Amazon in the robotics arena? Are they very much ahead of the curve on this? Do you think that they will get to it faster than, for example, you know, I know Tesla has had robot emissions. They're much bigger. But it seems to me like Amazon actually has an application to use the robots in a way that's not like actually developing humanoids, for example.

16:42Yeah. You know, I think that Amazon is actually very advanced in robotics. They've been at it for like almost a decade. But I think unlike some of these other companies, they're not really trying to sell their robots as like enterprise offerings to other businesses. They want to use them for themselves so far. So I think it's kind of hard to like compare them in some ways to Tesla, which I think is trying to be sort of like a robotics powerhouse where it's like, I think Amazon kind of sees robotics as more of accentuating its existing e-commerce business. And they use them in other ways, like an AWS data centers, for example.

17:20But yeah, remains to be seen. I mean, they definitely are actually quite, I think a lot more advanced than people know about. Well, I mean, but to your point, I mean, I don't know the exact history of AWS, but I mean, that seemed to have been a project that I don't know if they ever thought they were going to sell it externally. But, you know, conceivably it was something that they did to help their own business first, and then they figured out that it could turn into a business down the line. So, one to watch. Yes, totally. Yeah, I mean, they like making money. Okay. All right. Well, thank you very much, Catherine.

17:53I appreciate you coming on. And that is Catherine Perloff, our Amazon reporter, here at The Information.

17:59Laura Mandaro:Aside from GTC, a number of big tech companies like Amazon and Meta made headlines this week with new revenue projections and new cloud deals. We also had more AI startups raising big funding rounds. This morning, the White House also released its new framework for AI legislation. This week on the Editor's Cut, we are going to break down all of that. I want to bring on our co-executive editor, Martin Peers, and our managing editor, Laura Mandero. Welcome to you both. It's great to have you back. Hiya, guys. Hiya, Tosh. Martin, what do you make of the National AI Legislative Framework out this morning?

18:35Aaron Holmes:You know, it's a very nice statement of ideals that I don't think means very much. I mean, basically, if you read this, it says, AI, we want parents to control AI for their children. We want creators to be protected. We want to encourage innovation. It doesn't mention they want AI to cure cancer and they want AI to end poverty, but you get the idea that that's what they're going for. But right at the end, they mentioned, oh, by the way, we don't want any of the states to regulate AI, which is what this is all about. And the problem they're going to have is that there's obviously a lot of unhappiness and and distrust among regular people about what AI will bring.

19:18Aaron Holmes:It's turning into a political issue for the White House. And you can see that when you have people like, you know, Ron Santas, the Florida governor out there really pushing for controls on AI. The White House is really taking the side of the tech industry. and I think the tech industry's interest in AI are not the same as regular people. The tech industry sees this as a way to make a huge amount of money. And so this, I think, this is just starting off what's going to be a fair amount of debate. So, Laura, every time the White House comes out with these types of executive orders, statements, frameworks, I mean, we've seen a couple of these, I think, in the AI era.

20:11We also saw them in the crypto era, you know, when the government was coming out saying, hey, here's how we're thinking about it.

20:17Laura Mandaro:I seem to remember that people in Silicon Valley, venture capitalists online, I mean, a lot of them, I think they get pretty excited about these things, right? They say, oh, they're, you know, we're getting on board with it. Yeah, I don't know what the reaction is to this one. I mean, I don't – it does not strike me that that is their target audience. I was reading this one and then comparing it to the executive order in December. And it's quite telling that the first couple points are about electricity and children. And it seems very geared towards the populace, maybe ahead of the midterm elections and speaking to the things that are on people's minds, particularly power prices.

21:00I mean, this is, you know, this is already was an issue last year. And I think with the oil shock will be maybe even more of an issue. So, you know, the specifics about how to regulate AI or not were lower down. I mean, they were there, obviously. And the administration does want to quash these state laws, which keep popping up. One interesting point is I think there's this dynamic around guardrails. I mean, some of the state laws have attempted to put some guardrails or even require certain models to get approval from the government, which definitely Silicon Valley does generally not like because they would say it would slow the pace of innovation.

21:46And that is all to prevent sort of catastrophic issues rising from these AI models. And the policy framework does kind of note, maybe we should have a guardrail to make sure there's not kind of censoring going on in the models. That is contrary to the ideology that the administration espouses. So there is a little bit of a dynamic as are some guardrails actually okay or not? But I didn't think that was the main point of this policy framework. It seemed very much addressed to the ordinary person who may blame AI for various things. And Martin, do you think it's effective in quashing those concerns for the ordinary person?

22:30Aaron Holmes:No, not at all. I mean, I think what people, I mean, people don't, you know, read these things and they certainly don't believe what they hear from the government. I think what will really, it's how people actually use AI and what they hear about the impact of it. I mean, everybody is concerned about jobs. That's obviously one of the biggest issues. And also there's the whole issue about AI making it easier for hackers to take over your bank accounts. That's the thing that I hear about. I think ordinary people are very concerned about all the issues with AI and how it will hurt them. And I think this is one of the issues for tech is that there's an enormous disconnect between regular people and the tech industry where ordinary people do not like the implications of AI.

23:25Well, okay, Laura, so Martin makes a good point here. I mean, ordinary people, they have issues with it. We just, and I mean, not even ordinary people. We saw that the companies themselves this week, Meta, we reported had the security incident with its own agent going rogue.