scripod.com
Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub

Latent Space: The AI Engineer Podcast

15 HOURS AGO
Latent Space: The AI Engineer Podcast

Latent Space: The AI Engineer Podcast

15 HOURS AGO
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

From the Bitter Lesson of AI scaling to the unsolved mysteries of protein folding, Google DeepMind’s Pushmeet Kohli and Biohub’s Sal Candido are rethinking what it takes to build AI that truly understands biology. In this special panel moderated by Brandon...

Highlights

This panel examines why AlphaFold was a landmark without being the end of protein biology. Pushmeet Kohli and Sal Candido discuss how to find useful biological scaling laws, choose between better models and better data, move from isolated proteins toward whole systems, and build AI that is both powerful and trustworthy.
01:14
The real work is finding the scaling law.
05:54
The problem comes first
08:30
Define the problem first
11:32
Scientific intuition can give models an unfair advantage
14:20
The next challenge is finding the information models need
16:57
AlphaFold did not solve all of protein dynamics
19:49
From spokes to whole bicycles
21:52
The challenge is extracting knowledge from the machine
24:47
Calibration matters more than perfect scores
27:34
Future models may explain AlphaFold
30:37
Start by asking what a 10x improvement would require

Chapters

Introduction: The Bitter Lesson for Biological Data
00:00
Finding Scaling Laws and the Right Data for Biology
01:00
DeepMind’s Bitter Lesson: Solving Problems vs. Scaling Models
04:54
AlphaFold, Data Limitations, and Building the Virtual Cell
06:50
Handcrafted Architectures vs. Scaling Compute
09:52
Good Data, Inductive Bias, and Model Design
11:53
Beyond Transformers: The Future of AI Architectures
13:39
Why AlphaFold Hasn’t Solved Protein Folding
14:38
Protein Dynamics, Design, and Cryo-EM
17:39
From Individual Proteins to Whole Biological Systems
19:09
Feynman’s Principle: Creating Without Understanding
20:43
The Hidden Knowledge Inside Protein Language Models
22:18
AlphaFold, Trustworthiness, and Interpretability
23:48
Could AI Understand Other AI Models Better Than Humans?
25:56
When Will AI Revolutionize Drug Discovery?
27:43
Why Curing All Disease Requires 10x Breakthroughs
29:47

Transcript

Brandon Anderson: great to be here. What an exciting morning. So many cool announcements. I think the future of bioscience is being announced right here. So this is the modeling session. We are all modelers. So, of course, it's natural for us to talk about...