NVIDIA VP Ming-Yu Liu: Cosmos 3, World Models, Kung Fu, What Jensen Taught Me
Ming-Yu Liu, VP of Research at NVIDIA, discusses the development of Cosmos 3, a world model for Physical AI, and shares insights on NVIDIA's culture of collaboration, ambition, and the company's approach to research and development. He emphasizes the importance of generalization in AI and the potential of the Cosmos model to advance Physical AI applications.
Summary
The interview features Ming-Yu Liu, NVIDIA's VP of Research, who reflects on his decade-long journey at the company and the development of the Cosmos 3 world model. He describes how NVIDIA’s research approach has evolved, especially within the context of Physical AI. Liu states that the future of AI relies on constructing robust models that generalize well beyond training data, a challenge central to the Cosmos project. He emphasizes that the mission of helping users succeed underpins their work, and this has fostered a culture of collaboration and low ego within NVIDIA. The conversation touches on Jensen Huang's influence, the importance of rapid iteration and transparency in research, and how the competitive landscape is shifting towards a collaborative ecosystem where companies can provide mutual benefits. Liu discusses the challenges of managing a large team and the organizational structures that allow for autonomy while maintaining alignment toward common goals. He concludes with optimism about potential breakthroughs in Physical AI, which he sees as a vital frontier in AI development.
Key Insights
- Ming-Yu Liu believes that the success of Physical AI is heavily dependent on generalization, which entails learning skills from limited signals and applying them in unseen scenarios.
- Liu emphasizes that the process of building Cosmos involved significant iterations and that the model's capabilities outpace any of the individual previous models combined.
- He argues that world models are tools for solving specific problems, and not all companies need to build their own models, indicating that the focus should be on creating value rather than just competing.
- Liu describes Jensen Huang's management style as one of deep involvement and support, stating that Huang often reads employees' emails and prioritizes understanding their concerns.
- He reflects on NVIDIA's goal to assist others in building Physical AI solutions by providing general foundational models, thus creating a mutually beneficial ecosystem.
Topics
Transcript
[0:00] This English translation was generated by AI and is for reference only. Hello, everyone. I'm Xiaojun. Today I'm in Shanghai, and my guest is Ming-Yu Liu, NVIDIA's VP of Research. Our conversation takes place one month after Jensen Huang's recent visit to China. Ming-Yu has always struck me as someone who dresses very plainly, carrying a somewhat worn crossbody bag, with a very humble demeanor. [0:31] But people inside NVIDIA told me he doesn't seem like a typical researcher. He's more like an engineering leader. Jensen Huang gave him an evaluation: GSD — getting shit done. Not long ago, Ming-Yu Liu led his team to release NVIDIA's world model, Cosmos 3. So in this interview, we talked about the…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Zhang Xiaojun Podcast
Anker / Steven Yang: Consumer Electronics Death & birth, The Third Category, Product Philosophy
The interview features Steven Yang, CEO of Anker, discussing the challenges and strategic shifts within the consumer electronics industry, particularly in response to AI advancements. He emphasizes the importance of creating tangible customer value and the company's evolution from a focus on charging products to more complex technologies, including AI and chips.
Yao Shunyu: Let Me Go a Little Crazy! Training Models at Anthropic & Gemini, Heroism Is Over
Yao Shunyu, a researcher who moved from Anthropic to Google DeepMind, discusses the current state of AI model development, the competitive landscape between major AI labs, and his personal journey from theoretical physics to AI research. He shares candid views on why individual heroism has ended in AI, the importance of reliability over brilliance, and his technical perspectives on pre-training, post-training, and long-horizon tasks.
Luo Fuli: OpenClaw, Agent Frameworks — The AI Paradigm Has Already Changed Dramatically!
Luo Fuli, head of Xiaomi's large model division, describes how her firsthand experience with OpenClaw over Spring Festival 2026 fundamentally changed her understanding of AI agent frameworks as a paradigm shift — not just a product. She explains how OpenClaw's open-source, sophisticated context orchestration enabled her team to dramatically accelerate research and model training, and outlines how this Agent era demands a new approach to model architecture, post-training, and organizational design.
A 4-hour Interview with Carina Hong: AI for Math, Lean, Proofs from The Book, and Intuition
Carina Hong (洪乐潼), a 24-year-old Chinese-born founder, discusses her company Axiom, which recently closed a $200M Series A at a $1.6B valuation, focusing on AI for Math using formal proof systems like Lean. She shares her journey from competitive math olympiads in Guangzhou to MIT, Oxford, and Stanford, before dropping out to build an AI theorem-proving system that achieved the first-ever perfect score on the Putnam mathematics competition.