scripod.com
What Comes After ChatGPT? The Mother of ImageNet Predicts The Future

Overview

Shownote

Highlights

Transcript

Chapters

Pins

What Comes After ChatGPT? The Mother of ImageNet Predicts The Future

The a16z Show

2025/12/05
The a16z Show

The a16z Show

2025/12/05
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

Fei-Fei Li is a Stanford professor, co-director of Stanford Institute for Human-Centered Artificial Intelligence, and co-founder of World Labs. She created ImageNet, the dataset that sparked the deep learning revolution.  Justin Johnson is her former PhD ...

Highlights

In this conversation, two leading AI researchers delve into the frontier of spatial intelligence in artificial systems, exploring how machines can move beyond language to understand and generate physical worlds. Their work challenges conventional approaches in deep learning and opens new pathways for modeling reality.
22:40
Pixels may be a more lossless representation than tokenized text in AI models
30:33
Marble generates 3D worlds from text or images and allows interactive editing in real time.
52:58
Newtonian laws are unlikely to emerge from LLMs because they operate at a different abstraction level than astrophysical data.
55:43
Transformers are natively models of sets; sequence order comes from positional embeddings.

Chapters

Why academia must lead the next wave of AI innovation
00:00
The hidden flaws in today’s neural networks and tokenization
11:44
How Marble builds interactive 3D worlds from text or images
25:13
Is spatial intelligence something language models will ever truly grasp?
44:36
What if transformers aren’t sequence models at all?
55:43

Transcript

Alessio Fanelli: 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, wh...