Talks

Can AI surpass physics? A case study in weather & climate modeling

NeurIPS AI for Science Workshop, San Diego — December 2025.

An invited talk on what the successes and challenges of AI-based weather and climate modeling tells us about the potential for replacing traditional physics-based approaches.

Can AI surpass physics? — NeurIPS 2025 talk

Google’s AI-based Weather and Climate Modeling with NeuralGCM

BuzzRobot, Palo Alto — October 2024.

An overview of NeuralGCM for a non-technical audience, and my perspective on the broader field of AI weather and climate modeling.

NeuralGCM — CLEX Seminar Series

ARC Centre of Excellence for Climate Extremes — 2024.

A research-audience version of the NeuralGCM talk, with more detail on the science and on how the model fits into the broader weather/climate modeling landscape.

▶ Watch on YouTube

Deep learning and differentiable simulations

Google — 2021.

A broader take on my AI-for-science research program: why differentiable simulations are a natural meeting point for machine learning and traditional scientific computing, and what we have learned from applying the approach to fluid dynamics, structural optimization, and quantum chemistry.

▶ Watch on YouTube

Introducing xray: N-D Labeled Arrays and Datasets

SciPy 2015.

The original Xarray launch talk. Why labeled multi-dimensional arrays make working with scientific data dramatically easier, and how Xarray builds on NumPy and Pandas.

▶ Watch on YouTube