scripod.com
After LLMs: Spatial Intelligence and World Models — Fei-Fei Li & Justin Johnson, World Labs

Overview

Shownote

Highlights

Transcript

Chapters

Pins

After LLMs: Spatial Intelligence and World Models — Fei-Fei Li & Justin Johnson, World Labs

Latent Space: The AI Engineer Podcast

2025/11/25
Latent Space: The AI Engineer Podcast

Latent Space: The AI Engineer Podcast

2025/11/25
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

Fei-Fei Li and Justin Johnson are cofounders of World Labs, who have recently launched Marble (https://marble.worldlabs.ai/), a new kind of generative “world model” that can create editable 3D environments from text, images, and other spatial inputs. Marbl...

Highlights

The conversation dives into the evolution of AI beyond language, exploring how spatial intelligence is reshaping the way machines understand and generate 3D environments. From foundational research to real-world applications, the discussion bridges decades of computer vision progress with a bold new direction in generative modeling.
09:12
Academia should focus on new and wacky ideas rather than training large models.
19:39
Pixels may be a more lossless and general representation of the world compared to tokenized representations used in LLMs.
22:26
A model can render realistic scenes without understanding the physical forces behind them.
41:14
Spatial intelligence complements linguistic intelligence and involves reasoning, understanding, moving, and interacting in space.
47:26
Most humans are born with the ability to link perception and motor movement, a capability that remains challenging for AI.
50:11
An LLM may predict accurate movement trajectories but not derive Newtonian laws.
57:58
We need intellectual fearlessness—this is a pioneering field demanding bold thinkers.

Chapters

How a Vision for Smarter Machines Led to World Labs
00:00
When Pictures Started Talking: The Rise of Vision-Language AI
14:19
Why Can’t AI Truly Understand How Things Fall?
22:26
Meet Marble: A Living Canvas for 3D Worlds
30:42
Seeing vs. Describing: What Language Misses About Space
44:15
Could an AI Ever Discover Newton’s Laws on Its Own?
50:11
Join the Next Wave of AI: Build Beyond Language
57:58

Transcript

Justin Johnson: I think the whole history of deep learning is, in some sense, the history of scaling up compute. Fei-Fei Li: When I graduated from grad school, I really thought the rest of my entire career would be towards solving that single problem, whi...