Descriptions:
Most of AI is pointed at language. Anima Anandkumar points it at the physical world, weather, fusion, materials, the systems physics writes down as equations, and finds the usual playbook breaks. There isn’t enough data, the resolution is brutal, and no transformer is big enough. Her way through is older than deep learning: build the structure of the world into the model. It already produces weather forecasts that rival the supercomputers on a single GPU.
Anima has spent two decades in AI, from the theory that predates deep learning, through scaling it at AWS and NVIDIA, and back to first principles. We get into neural operators, Fourier layers, and the spherical harmonics behind FourCastNet 3, then follow the same ideas into fusion reactors, chip design, and formal verification, before closing on her seat on the UN Scientific Advisory Board and why AI for science shouldn’t be regulated like a chatbot.
CHAPTERS
0:00 Cold open and intros
3:48 Anima’s thesis: AI for science, not language models
6:44 TorchLean: proving what a neural network will do
12:06 Why physics-informed nets fail, and how neural operators fix it
16:08 What a neural operator actually is
19:41 Fourier neural operators, and why Fourier
26:53 Old math meets deep learning, and why scaling breaks
32:39 Weather, and the 2021 skeptics
36:13 One model for weather and climate: assume a sphere
42:40 Why the physical world is forgiving
46:19 Modeling across scales, from atoms to planets
50:33 The data, ECMWF, and calling Hurricane Lee early
57:48 Long rollouts, and why the sphere stays stable
1:04:06 Fusion: a plasma digital twin
1:07:59 Career full circle: theory, scale, and back to principled
1:13:02 A foundation model for physics, and inverse design
1:18:13 The UN, regulating AI for science, and a call to action
LINKS
Papers
FourCastNet: https://arxiv.org/abs/2202.11214
Neural Operators: https://arxiv.org/abs/2108.08481
Fourier Neural Operator: https://arxiv.org/abs/2010.08895
FourCastNet 3: https://arxiv.org/abs/2507.12144
TorchLean: https://arxiv.org/abs/2602.22631
Lean: https://lean-lang.org
Anima Anandkumar
Caltech: https://www.eas.caltech.edu/people/anima
UN Scientific Advisory Board: https://www.caltech.edu/about/news/anima-anandkumar-appointed-to-un-scientific-advisory-board
The show
Latent Space: https://www.latent.space
