Fei-Fei Li (World Labs) — On Creati...

1. Key Themes

Spatial Intelligence Is the Next Frontier After LLMs

Fei-Fei Li argues the entire AI industry is still fixated on language, while the more fundamental capability — understanding and interacting with the physical world — remains largely untapped. World Labs was founded specifically around this bet, with a $1 billion raise to prove it.

"This is just like LLMs. I think there will be many companies doing incredible work in world models." 00:05:26

"Animal intelligence starts with seeing and moving in the physical world. That evolution began with us as animals knowing what the world is, knowing who we are, knowing how to move around it, interact with it." 00:00:18

A Functional Taxonomy of World Models — Three Distinct Types

Fei-Fei Li makes the non-obvious point that "world model" is an overloaded term being used to describe three fundamentally different things, which muddies competitive analysis and investment theses. She defines them precisely.

"Right now there are three ways of calling world models when it comes to spatial intelligence. One is what I call a renderer... The second kind of world model is what we call a planner... The third kind which I think is the linchpin of the three is a simulator — it actually is consumed by humans as well as machines, is trying to respect the structure, the physics and the dynamics of the world and really simulate the 3D and 4D information." 00:05:26

World Labs is explicitly targeting the simulator layer, which she believes is the critical path to unlocking the other two.

The "ChatGPT Moment" for Spatial Intelligence Will Be Professional, Not Consumer

Unlike LLMs which had a viral consumer moment, spatial intelligence adoption will likely flow first through professionals — designers, engineers, roboticists, researchers — before (if ever) reaching mainstream consumers.

"The kind of applications we are talking about tends to be first going to the professionals. Professionals, professional creators, professional designers, professional developers, professional researchers and engineers who use it for robotics and industrial design and all that. So maybe we will not necessarily have a consumer moment." 00:03:19

Robotics Investment Is Dramatically Underfunded Relative to the Opportunity

Li pushes back on the narrative that $6 billion into humanoids is excessive hype. Her view is that the number is far too small given the comparable investments in self-driving and LLMs, and that robotics will be among the most important industrial revolutions in human history.

"Robotics is going to be one of the most important revolution in human industrialization. Six billion dollars is too small. If you look at self-driving cars investment, if you look at language models investment, it took way more than $6 billion." 00:08:14

AI Safety Theater vs. Real Safety Work

Li takes a measured, non-theatrical position on AI safety — neither doom nor utopia — and grounds her view in specific, real-world deployment evidence she witnessed firsthand in healthcare settings.

"Doctors are using AI to help them with charting. Radiologists are using AI to assist them reading the MRI and the CT scans... Safety measures are happening, but there needs to be more in the right way, in a scientifically grounded way. And that's the conversation that should be taking place instead of what you say, the theater." 00:11:29

The Public Discourse Vacuum Is Fueling Anxiety More Than AI Itself

Li identifies the polarized extremes of doomerism vs. utopianism as the actual cause of public fear — not AI's capabilities. The real problem is that ordinary people have nowhere to turn for nuanced, grounded answers.

"A lot of this sentiment happens when there is a vacuum of thoughtful public discourse. Right now the oxygen, the air is all sucked into the polarized extreme of doomerism or total utopian. And when hype takes all the oxygen in the room, that void fuels the kind of anxiety." 00:13:31

Education Must Be Fundamentally Restructured Around AI

Li argues this is one of the most important opportunities of the next decade — not incremental reform but a wholesale change in how we evaluate and teach, given AI now surpasses average humans on standardized tests including math Olympiads.

"When AI can do better than average human, it's not about humans are bad, it's about we need to change the education system. We need to change how we evaluate. We need to change the way we empower teachers to teach." 00:15:17

Refusing to Engage with AGI Framing as a Scientific and Strategic Choice

Li explicitly distances herself from the AGI narrative — not from ignorance but from a principled view that it's an ill-defined term that distracts from building real technology with real impact.

"I don't engage with the term AGI because the founding fathers of artificial intelligence as a scientific field had this dream of thinking and doing machines. And that is a scientific quest. And that quest has been my lifelong career... I'm okay people call it whatever they want. They can call it an apple. That's fine." 00:17:07

2. Contrarian Perspectives

The Simulator Layer — Not the Renderer or Planner — Is the Real Value Creation Layer

Most competitors and investors are focused on either video generation (renderers, visually impressive) or robotics planners. Li's contrarian claim is that the unglamorous "simulator" tier in between is actually the linchpin that unlocks both — and that's what World Labs is building.

"The third kind which I think is the linchpin of the three is a simulator... a simulator could become a renderer, the simulator could become a planner, but this layer is a huge critical path in my opinion to unlock spatial intelligence." 00:06:26

$6 Billion Into Robotics/Humanoids Is Underfunding, Not Overfunding

The mainstream narrative frames humanoid robot investment as hype-driven and excessive. Li inverts this completely — comparing it to the scale of capital that went into self-driving and LLMs — and suggests it will require far more.

"Six billion dollars is too small. If you look at self-driving cars investment, if you look at language models investment, it took way more than $6 billion." 00:08:14

Language Models Have a Fundamental Ceiling That Words Cannot Cross

In a world where LLMs are treated as general-purpose intelligence, Li draws a sharp line: there are entire categories of human capability that language will structurally never address.

"Can words put down fires? Can words cook an omelet?" 00:01:58

There May Be No ChatGPT Moment for Spatial Intelligence — And That's Fine

The prevailing assumption in AI is that breakthrough technologies announce themselves through viral consumer adoption. Li suggests spatial intelligence may not follow this pattern, arriving instead through professional verticals first — a more durable but less visible path to dominance.

"Maybe we will not necessarily have a consumer moment." 00:04:15

3. Companies Identified

World Labs

Spatial intelligence startup co-founded by Fei-Fei Li. Building large world models with a focus on the "simulator" layer — models that respect physics, geometry, and dynamics to enable both human and machine consumption. Raised $1 billion. NVIDIA is an investor.

"We are all in in spatial intelligence, and the means to spatial intelligence is building a large world model." 00:00:18

NVIDIA

Chip and AI infrastructure giant. Mentioned as both a competitor (Cosmos world model) and an investor in World Labs — a notable dual role.

"NVIDIA has its own world models, Cosmos. NVIDIA is also one of your investors." 00:04:31

Anthropic

AI safety-focused LLM company. Mentioned in the context of AGI timeline claims by its CEO.

"Anthropic CEO Dario Amodei has suggested AGI is two to three years out." 00:16:45

Google DeepMind

Mentioned for Project Genie (world model work) and Demis Hassabis's "foothills of the singularity" framing.

"Google shipped Project Genie." 00:04:31

Stanford Hospital

Cited as a real-world deployment environment where AI is actively being used in clinical settings — charting for doctors, radiology AI for CT and MRI reading.

"I just came from the hospital... I was just in Stanford Hospital looking at where AI is already being used." 00:10:34

4. People Identified

Fei-Fei Li

Co-founder and CEO of World Labs; former Stanford AI Lab director and Google Cloud AI chief; widely called the "godmother of AI" for her foundational ImageNet work. She is betting her next chapter on spatial intelligence as the frontier beyond LLMs.

"We really had a head start and understanding that this is going to be the next frontier of AI." 00:04:49

Yann LeCun

Chief AI Scientist at Meta (referred to as having left Meta to work on world models). Mentioned as a signal that the world model thesis is gaining institutional credibility.

"In the last six months, Yann LeCun left Meta to work on world models." 00:04:31

Dario Amodei

CEO of Anthropic. Mentioned for his two-to-three year AGI timeline prediction, which Li pointedly declines to engage with.

"Anthropic CEO Dario Amodei has suggested AGI is two to three years out. We'll get there by scaling the current paradigm." 00:16:45

Demis Hassabis

CEO of Google DeepMind. Mentioned for his "foothills of the singularity" framing — another AGI-adjacent claim Li sidesteps.

"Demis Hassabis says we're at the foothills of the singularity." 00:16:45

Eric Schmidt

Former Google CEO. Cited as a symbol of the AI backlash — getting booed at a college graduation — illustrating the real public sentiment gap between AI leaders and ordinary people.

"The former Google CEO Eric Schmidt getting booed at a college graduation." 00:12:18

5. Operating Insights

Publish a Taxonomy When Your Category Term Gets Co-opted

When a high-value term becomes overloaded and competitors begin using it loosely to describe inferior or different products, publish a clear definitional framework immediately. Li did exactly this — releasing a blog post the day before this interview defining the three types of world models — to own the narrative and differentiate World Labs before confusion calcified.

"24 hours ago, we kind of got fed up that the word world model has been so confusing and being used in so many different ways that we actually put out a blog just explaining what a functional taxonomy of world model is instead of mushing everything together." 00:05:26

Anchor Your Company's Origin Story to a Multi-Hundred-Million-Year Frame

When competing against incumbents with vastly more resources (Google, NVIDIA, Meta), your differentiation cannot be features — it must be conviction and depth of vision. Li anchors World Labs not to a market trend but to 500 million years of evolutionary biology, making the competitive moat feel intellectual and civilizational rather than tactical.

"The case for us is a 500 million year story, is that animal intelligence starts with seeing and moving in the physical world." 00:00:18

Target Professionals First When Consumer Virality Isn't the Natural Path

For deep technology products where the application complexity is high, resist the pressure to manufacture a consumer moment. Design the GTM around professionals who will derive immediate, measurable value — their credibility and word-of-mouth is more durable than viral adoption.

"The kind of applications we are talking about tends to be first going to the professionals." 00:03:19

6. Overlooked Insights

The Personal AI Use Case for Healthcare Navigation Is Already Here — and Underappreciated

In passing, while explaining she had rushed from a hospital to the panel, Li drops a single throwaway sentence that is actually a significant signal about a real, high-frequency consumer use case for LLMs that is already happening organically and at scale.

"I got this long radiology report last night and the first thing I did is send it to an AI so that they can help me to explain it." 00:11:29

This is a Stanford AI professor — not a regular consumer — defaulting to AI as the first-line interpreter of a complex medical document. If the most technically sophisticated users are already doing this reflexively, the consumer healthcare AI navigation market (medical report translation, pre-visit prep, treatment decision support) is likely far closer to mainstream adoption than the industry is pricing in. This points toward a large, under-invested wedge: AI tools built specifically for patients and family members navigating the healthcare system — not for doctors or hospitals, but for the people receiving care.

The "Spatial Intelligence for Creativity" Wedge Is Broader Than Robotics and Likely Faster to Monetize

Throughout the interview, the robotics and industrial applications dominate the framing. But Li briefly names a consumer creativity use case — interior design, home design — almost as an aside, that could represent a much faster path to revenue and product-market fit than waiting for robotics infrastructure to mature.

"I would love to design my home in a much easier way and just change the color of the curtain with a click." 00:04:15

"People design. People, whether we're designing interior space, we're designing machines, we're designing homes, we're designing stories. So much of that is beyond words." 00:01:58

The consumer and prosumer design market — interior design, architecture visualization, 3D product design — represents a near-term, high-willingness-to-pay wedge for world model technology that does not require solving robotics or industrial simulation. Companies building spatial AI tools for designers and homeowners could be early, high-margin distribution channels for the underlying simulator infrastructure World Labs is developing. This category is barely being discussed relative to the robotics hype cycle.