The Godmother of AI on jobs, robots & why world models are next | Dr. Fei-Fei Li

Nov 16, 2025 Episode Page ↗
Overview

Dr. Fei-Fei Li, the "godmother of AI" and co-creator of ImageNet, discusses AI's history, its impact on humanity, and the future with world models. She introduces Marble, the world's first large world model, enabling creation and interaction with 3D environments.

At a Glance
19 Insights
1h 19m Duration
12 Topics
6 Concepts

Deep Dive Analysis

AI's Impact on Humanity and Personal Responsibility

Brief History of AI: From Early Days to AI Winter

The Genesis and Breakthrough of ImageNet

The Rise of Modern AI and Deep Learning

Defining AGI and Future AI Innovation Needs

Introduction to World Models and Spatial Intelligence

Challenges and the 'Bitter Lesson' in Robotics

Introducing Marble: The First Large World Model Product

Applications and Use Cases of Marble

Fei-Fei Li's Founder Journey and Career Advice

Stanford's Human-Centered AI Institute (HAI) Mission

The Universal Role of People in AI's Future

AI Winter

A period from the late 1980s through the early 2000s when public interest, funding, and belief in AI significantly declined because early promises of AI were not met. During this time, some tech companies even avoided using the term 'AI' due to its negative connotation.

ImageNet

A massive dataset of 15 million carefully curated images categorized into 22,000 object concepts, created by Fei-Fei Li and her students. Its open-sourcing and accompanying challenge in 2012 provided the 'big data' necessary to train the first successful neural network algorithms, sparking the deep learning revolution.

Deep Learning (Modern AI)

The current paradigm of AI development, characterized by the combination of large datasets, neural network algorithms, and powerful GPUs. This approach, pioneered by the ImageNet breakthrough, has enabled significant advancements in areas like object recognition and conversational AI.

World Model

An AI model designed to understand and generate 3D spatial environments, allowing users to create, interact with, and reason within immersive worlds. It extends AI beyond language-centric tasks to encompass visual and spatial intelligence, crucial for applications like robotics and virtual reality.

Bitter Lesson (Richard Sutton)

A concept in AI stating that simpler algorithms, when combined with vast amounts of data and compute, often outperform more complex, hand-engineered models. While influential, its direct application to robotics is challenged by difficulties in data acquisition and the physical nature of robots.

Human-Centered AI

A guiding framework for AI development and application that prioritizes human benevolence, dignity, and agency. It emphasizes that AI should be inspired by people, created by people, and most importantly, impact people positively, requiring responsible individual and societal engagement.

?
How will AI impact humanity?

AI is a double-edged sword; its future impact is 'up to us' as responsible individuals and society, and it is fundamentally a net positive technology for humanity, making lives and work better and building civilization.

?
What was the state of AI before ImageNet?

Before ImageNet, the field was in an 'AI winter,' with limited public interest and funding, and researchers were primarily focused on mathematical models like neural networks but lacked the large datasets needed for effective training.

?
What was the key breakthrough that led to modern AI?

The key breakthrough was the combination of ImageNet's big data, neural network algorithms, and powerful GPUs, demonstrated in the 2012 ImageNet challenge, which significantly advanced object recognition and became the 'golden recipe' for modern AI.

?
Are we close to achieving Artificial General Intelligence (AGI) with current approaches?

Fei-Fei Li views AGI as more of a marketing term than a scientific one, but believes we are still far from achieving the full goals of AI, as current models cannot perform complex reasoning, abstraction, or emotional intelligence tasks that humans easily accomplish.

?
Why is the 'bitter lesson' alone not likely to work for robots?

The bitter lesson (simpler models with more data win) faces challenges in robotics because it's harder to acquire diverse 3D action data for training, and robots are physical systems requiring not just brains but also bodies and mature application scenarios, making their development a longer, more complex journey.

?
What are some practical applications of the Marble world model?

Marble is being used for virtual production in movies (significantly cutting production time), game development, robotic simulation (creating diverse synthetic training environments), and even psychological research to generate immersive scenes for studying psychiatric patients' responses.

?
How does Marble differ from existing video generation models?

Marble focuses on generating genuinely 3D, navigable, and interactable worlds with underlying spatial intelligence, allowing users to reason and move within them, rather than just passively watching 2D video outputs like other video generation models.

?
Does everyone have a role in the future of AI?

Yes, everyone has a role in AI, regardless of their profession, as AI is fundamentally about people, and human dignity and agency should be at the heart of its development, deployment, and governance, encouraging individuals to embrace AI as a tool for their unique contributions.

1. Center Human Dignity in AI

Ensure that human dignity and agency are at the core of AI’s development, deployment, and governance, preventing technology from diminishing human value.

2. Shape AI’s Future

Recognize that AI’s development and impact are ultimately determined by human choices and actions, emphasizing individual and societal responsibility.

3. Act Responsibly with Technology

Understand that every technology, including AI, is a double-edged sword, requiring responsible societal and individual actions to prevent negative outcomes.

4. Be a Responsible AI Participant

Act as a responsible individual in all aspects of AI development, deployment, and application, caring deeply about its impact on individual life, community, and future generations.

5. Care About AI’s Impact

Recognize that AI will impact your individual life, community, society, and future generations, making it crucial to care about its development and application.

6. Cultivate Intellectual Fearlessness

Embrace intellectual fearlessness and courage when pursuing new endeavors or seeking to make a difference, as it’s essential for creating something novel where others haven’t ventured.

7. Prioritize Passion & Mission

When considering career opportunities, focus on your passion, alignment with the mission, and belief in the team, rather than over-analyzing every minute detail of the job.

8. Avoid Overthinking Downsides

When pursuing ambitious goals, avoid getting bogged down by overthinking every possible negative outcome, as this can hinder progress and courage.

9. Innovate Beyond Scaling Laws

While scaling with more data, GPUs, and larger models is important, actively seek and pursue new innovations rather than assuming current approaches will solve all future challenges.

10. Identify Overlooked Ingredients

Look for critically overlooked components in a problem (e.g., big data for AI) and ambitiously pursue their creation and open-sourcing to accelerate progress in a field.

11. Work with Bright, Fast Teams

To bring new technology to life quickly, prioritize working with the brightest technologists and foster an environment that allows for rapid progress.

12. Embrace Audacious Questions

Don’t shy away from using ambitious or “audacious” terms for your field or work, especially if they represent fundamental questions in science and technology.

13. Utilize World Models

Leverage world models like Marble to create, reason within, and interact with infinitely explorable 3D environments, enabling applications in design, gaming, and simulation.

14. Embrace AI as Creative Tool

If you are an artist or storyteller, embrace AI tools like Marble to enhance your unique storytelling, leveraging them to express your voice in innovative ways.

15. Augment Healthcare with AI

Explore and implement AI technologies, such as smart cameras or robotic assistance, to augment and support healthcare workers, addressing issues like overwork and an aging society’s needs.

16. Participate in AI Governance

As a citizen, regardless of your profession, actively participate in your community and voice your opinion on how AI is used and applied to ensure it benefits everyone.

17. Use Figma Make for Prototypes

Use Figma Make to quickly turn ideas and designs into fully functional prototypes or apps with just a few prompts, leveraging existing design building blocks.

18. Show, Don’t Just Tell

Instead of spending time explaining product vision, use tools like Figma Make to create code-backed prototypes and apps to visually demonstrate your ideas.

19. Streamline HR with JustWorks

Utilize JustWorks for simplified HR, automated payroll, premium benefits, and international hiring support to effectively manage your team across different locations and needs.

There's nothing artificial about AI, it's inspired by people, it's created by people, and most importantly, it impacts people.

Fei-Fei Li

I do believe technology is a net positive for humanity, but I think every technology is a double-edged sword.

Fei-Fei Li

AI is a field of, at this point, 70 years old. And we have gone through many generations. Nobody, no one could have gotten here by themselves.

Fei-Fei Li

AGI is more marketing term than a scientific term.

Fei-Fei Li

The more I work in AI, the more I respect humans.

Fei-Fei Li

I don't overthink of all possible things that can go wrong because that's too many.

Fei-Fei Li
70 years
Age of AI as a field Dating back to the 1950s
2000
Fei-Fei Li's entry into AI field (PhD start) Year her PhD began at Caltech
15 million
ImageNet curated images Images on the internet
22,000
ImageNet object concepts Taxonomy of concepts
2012
Year of deep learning breakthrough (ImageNet challenge) When Toronto researchers used ImageNet with neural networks and GPUs
3 years ago
Approximate time ChatGPT became publicly known From the time of the podcast recording
2016
Year AI was considered a 'dirty word' by some tech companies Some companies avoided using the term AI
2017-ish
Approximate year companies started calling themselves AI companies Beginning of this marketing trend
20 years
Time from self-driving car prototype to today's Waymo Since Stanford's car won the DARPA challenge in 2005/2006
20 watts
Human brain power consumption Less than a typical light bulb
40X
Production time reduction for virtual production using Marble Observed by a virtual production company collaborating with Sony
30ish people
World Labs team size Predominantly researchers or research engineers
2018
Stanford's Human-Centered AI Institute (HAI) founding year Co-founded by Fei-Fei Li and other faculty
6-7 years
Age of HAI Institute Since its founding in 2018