Moonlake: Multimodal, Interactive, and Efficient World Models — with Fan-yun Sun and Chris Manning

Latent Space · Beginner ·📄 Research Papers Explained ·8h ago
We’ve been on a bit of a mini World Models series over the last quarter: from introducing the topic with Yi Tay, to exploring Marble with World Labs’ Fei-Fei Li and Justin Johnson, to previewing World Models learned from massive1 gaming datasets with General Intuition’s Pim de Witte (who has now written down their approach to World Models with Not Boring), to discussing the Cosmos World Model with with Andrew White of Edison Scientific on our new Science pod, to writing up our own theses on Adversarial World Models. Meanwhile Nvidia, Waymo and Tesla have published their own approaches, Google …
Watch on YouTube ↗ (saves to browser)

Chapters (13)

Benchmarking Gets Hard
0:47 Meet Moonlake Founders
1:26 Why Build World Models
3:12 Structure Not Just Scale
5:37 Defining Action Conditioned Worlds
7:32 Abstraction Versus Bitter Lesson
14:39 Language Versus JEPA Debate
20:27 Reasoning Traces And Rendering Layer
37:00 Gameplay Over Graphics
38:02 Fiction Rules And World Tweaks
39:15 Code Engines Beat Learned Priors
41:10 Diffusion Scaling Limits
43:23 Symbolic Versus Diffusion
Final Step: Take the Quiz & Claim Your Certificate
Next Up
Final Step: Take the Quiz & Claim Your Certificate
Analytics Vidhya