Feb 5, 2025 · 1h 0m · big-technology
NVIDIA VP Rev Lebaredian Talks Plan To Build AI That Understands The Real World
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
NVIDIA VP of Omniverse and Simulation Technology Rev Lebaredian joins the Big Technology Podcast to discuss NVIDIA's push into physical AI, world foundation models, and robotics. He outlines how multimodal simulation platforms like Project Cosmos bridge the gap between digital models and real-world physical dynamics to power autonomous machines across a multi-trillion-dollar economy.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Alex holds 23.7% of the talking time here. How this is scored →
Alex as informed peer 5.1 Guest teaching 5.5 Guest disagreement 1.3 Alex pushing back 1.9
speaking balance: gold is Alex, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:10 Rebuttal on video models lacking true physics
Rev directly rejects the premise that video generation models possess genuine physics understanding, pointing out fundamental flaws like object permanence failures.
Hardest push from Alex ▶ 25:27 Challenging the need for simulation over video generationAlex challenges Rev with a detailed multi-part question, questioning why expensive simulation stacks are needed when modern video generators already appear to understand physical dynamics.
Biggest teaching moment ▶ 37:31 Educating on NVIDIA full-stack software identityRev dismantles the common external misconception of NVIDIA being just a chip manufacturer by explaining that the vast majority of engineers are software developers building full-stack platforms.
Alex holds their own ▶ 6:00 Integrating Yann LeCun physics thought experimentAlex showcases deep technical familiarity by citing Yann LeCun's paper-drop test and connecting it directly to NVIDIA's world model thesis.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Alex as informed peer | Guest teaching | Guest disagreement | Alex pushing back | Why |
|---|---|---|---|---|---|---|
| Jevons Paradox and Computing Demand | 5 | 6 | 1 | 2 | Alex brings up Jevons Paradox in the context of DeepSeek and compute demand. Rev explains the economics and historical computing precedent of rendering and graphics at NVIDIA, expanding on the economic principle. | |
| AI Efficiency Gains and Full-Stack Performance | 7 | 6 | 1 | 2 | Alex demonstrates deep context by referencing his previous interview with Yann LeCun regarding physical common sense and dropping paper. Rev agrees and details how LLMs extract grammar rules versus physical world rules. | |
| Multimodal Foundations for Robot Brains | 5 | 5 | 1 | 2 | Alex clarifies whether real-world knowledge is meant for LLMs or physical robot foundations. Rev clarifies the multi-sensory and 3D simulation inputs required for robot foundation models. | |
| Project Cosmos and NVIDIA Omniverse | 4 | 6 | 0 | 1 | Alex asks about the newly announced Cosmos project. Rev details a ten-year progression from Omniverse physics simulation to tokenizers and open-weight models. | |
| Industry Applications and Ecosystem Adoption | 4 | 4 | 0 | 1 | Alex asks which developers will use Cosmos. Rev explains the broad scope spanning robotics, smart city infrastructure, and sensor coordination. | |
| Multimodal Ground Truth and Learning Mechanics | 5 | 5 | 1 | 2 | Alex inquires how textual knowledge integrates with physical perception. Rev uses childhood sensory developmental analogies to explain how multimodal association operates in AI training. | |
| Synthetic Simulation versus Video Generation Flaws | 7 | 7 | 3 | 4 | Alex mounts a detailed counterargument citing current video generation models that already display apparent physics understanding and infinite scenario generation. Rev counters firmly by highlighting major physics violations like object permanence and rendering errors. | |
| Visual Plausibility versus Physics for Robotics | 5 | 6 | 2 | 2 | Alex notes how convincing video effects can deceive viewers. Rev differentiates visual plausibility in Hollywood entertainment from the ground-truth physical simulation required for safe robotics. | |
| Scaling Laws and AI Comprehension of Physics | 6 | 5 | 1 | 2 | Alex connects Demis Hassabis and Yann LeCun's views on scaling laws and physics comprehension. Rev reflects on the surprising power of scaling laws while noting models remain on an exponential trajectory. | |
| NVIDIA's Full-Stack Identity Beyond Chips | 5 | 6 | 2 | 2 | Alex questions the outside perception that NVIDIA is merely a hardware chip designer. Rev educates on NVIDIA's software-heavy engineering makeup and full-stack model development history with Megatron. | |
| Labor Shortages and Humanoid Robotics with GR00T | 5 | 6 | 2 | 3 | Alex presses on labor displacement from humanoid robots like Project GR00T. Rev reframes the issue from job replacement to filling acute demographic labor shortages across industrial and supply chain sectors. | |
| Deployment Horizons from Factories to Healthcare | 5 | 5 | 2 | 2 | Alex explores specialization boundaries between human and robotic nurses. Rev outlines adoption timelines, arguing factories will lead while domestic kitchens and empathetic eldercare will remain human-dominated. | |
| The Impact of World Models on Hollywood CGI | 4 | 6 | 2 | 1 | Alex asks if Hollywood CGI will transition fully into AI generation. Rev draws on his personal visual effects background to explain why generative world models will radically lower production costs. | |
| Autonomous Warfare, Ethics, and Governance | 5 | 5 | 1 | 2 | Alex raises the dark ethical implications of robotic warfare. Rev acknowledges risks while citing historical arms control frameworks like nuclear treaties as precedents for global AI governance. | |
| Twenty-Three Years at NVIDIA and Corporate Culture | 4 | 4 | 0 | 1 | Alex asks for cultural insights from Rev's 23-year tenure at NVIDIA. Rev reflects on high retention rates, life's work culture, and the company's long-term bet on CUDA. |
Statements from this episode (24)
Lebaredian: Making AI more efficient will increase overall compute demand
“I think intelligence is something that is probably the most endless of all, all computing problems. If we can throw more compute at the problem, we can make more intelligence and do it better and better. So making, making AI more efficient, Will just increase …”
Lebaredian: NVIDIA achieved a million-fold AI compute gain over 10 years
“Over the last 10 years, we've had essentially a million x performance increase.”
Lebaredian: DeepSeek efficiency gains fit NVIDIA's decade-long progress curve
“So what we've seen here with DeepSeq is, is a great advancement that's on that same curve that we've been on for a decade now.”
Lebaredian: Future AI's greatest value will come from physical world interaction
“And our thesis is that from all the AIs we're going to create into the future, the most valuable ones are going to be the ones that can interact with our physical world.”
Lebaredian: Physical economy represents $100T annually compared to $2T-$5T for IT
“The world of knowledge, information technology is somewhere between two to five trillion dollars a year, but everything else, transportation, manufacturing, supply chain, warehouse and logistics creating drugs, all the stuff in the physical world, that's about…”
Lebaredian: NVIDIA trains physical AI models with video and 3D simulation
“We're not just going to describe with words, how, what happens when you drop a piece of paper. We're going to give these models other senses. During the learning process. So they'll, they'll watch watch videos of paper dropping. We can also give it more, more …”
Lebaredian: Multimodal world models will serve as brains for physical robots
“And so what we'll end up with is a world foundation model that was trained on many different modes of data, essentially different sense, senses. It can see, it can hear, It can touch and feel and do, do many of the things we can do, or many things other animal…”
Lebaredian: NVIDIA has worked towards Cosmos for about a decade
“We've been working towards Cosmos For probably about 10 years, we envisioned that eventually this new technology that had formed with deep learning, that that was going to be the critical technology necessary for us to create robot brains.”
Lebaredian: Video AI models fail basic physics like object permanence
“One of the basic things is object permanence. If you direct the video to move the camera, point away, and come back, Objects that were there at the beginning of the video are no longer there or they're different, right? And so that is such a fundamental violat…”
Lebaredian: Physical AI models only have 5% to 10% of needed understanding
“That all being said, I think we're going to rapidly get better and better. So, so the models today have an amazing amount of knowledge about the physical world, but they're maybe at like five, 10% of what they should understand.”
Lebaredian: Visual-Level Simulation Is Insufficient for Training Robot Brains
“That level of simulation is, is not sufficient. For building physical AI that are gonna be the underpinnings or the fundamental components of a robot brain.”
Lebaredian: NVIDIA Cosmos Is a Different Class of AI From Video Generators
“What we're doing with cosmos it's, It is really a different class of AI than video generators. You can use it to generate videos, but the purpose is different. It's not about generating beautiful imagery or interesting imagery as for art. This is about simulat…”
Lebaredian: AI Physics Understanding Will Unlock Most Robotics Applications in a Few Years
“At this point, I believe in a few years, we're going to get to a level of a physics understanding with our AIs that are, that's going to unlock, you know, the majority of the applications we need to apply them in, in robotics.”
Lebaredian: Majority of NVIDIA's engineers are software engineers
“The majority of our employees are engineers, and the majority of those engineers are software engineers.”
Lebaredian: ChatGPT would not exist without NVIDIA's early LLM software work
“Had we had not done that, I don't think we would have had ChatGPT.”
Lebaredian: Automotive factory workforces in Detroit and Germany face aging crises
“If you go to an automotive factory in in Detroit or in Germany, go look around. Most of the factory workers are aging and they're quickly retiring. And these CEOs that I'm talking to, their biggest concern is all of that knowledge they have on how to operate t…”
Lebaredian: Truck driver shortages require autonomous driving to maintain supply chains
“There's not enough truck drivers in the world to go deliver all the stuff that's moving around in our supply chains. We can't hire enough of them and there's less and less young people that want to do that job every year. So we need to have self-driving trucks…”
Lebaredian: Humanoid robots will take off first in industrial sectors
“We believe that the first place we're going to see general purpose robots like the humanoid robots really take off is in the industrial sector because of two things.
One, the demand is great there because we have the shortage of workers.
and also because
…”
Lebaredian: Homes will be the last place humanoid robots appear
“I think the last place we're going to start seeing humanoids show up is in our homes.”
Lebaredian: Domestic humanoid robots will adopt in Japan before Germany
“They'll probably show up in a kitchen in somebody's home in Japan before they show up in a kitchen in somebody's home in, in Munich, Germany.”
NVIDIA VP: Film studios will adopt AI world models for faster, cheaper VFX
“What we're building with AI, with generative AI, and particularly with world foundation models, that once we get to the point where they really understand, ah, the depth of the physics that they need to produce something like Planet of the Apes, once we have t…”
Lebaredian: History of weapons treaties gives reason for optimism on AI warfare
“We've done the same with biological weapons and chemical weapons. Largely, they haven't been used, even though the technologies existed there. And so I think that's a that's a good indicator of what's, how, how, how we should deal with this new technology, thi…”
Lebaredian: Over 650 NVIDIA employees reached a 20-year tenure
“When he got to 20, there were more than 650 people that were at twenty-year. Now, earlier I had said, when I joined the company, there were about a thousand people. So this means that most of the people That were there when I started at NVIDIA were still there…”