Into the Impossible With Brian Keating The Computer Expert That Just Solved AI’s TOUGHEST Challenge (ft. Rose Yu)
10 snips
Aug 26, 2025 
Guest Rose Yu is a computational physicist from UC San Diego, known for her groundbreaking work in AI, including traffic forecasting and pandemic modeling. In this engaging discussion, she reveals how AI could become a true partner in scientific discovery, generating novel hypotheses and enhancing modeling accuracy. Rose dives into the challenges of AI in academia, its potential to innovate beyond traditional methods, and even questions whether AI can experience emotions. This conversation is a fascinating look at the future of AI in science.
AI Snips
Chapters
Books
Transcript
Episode notes
AI Can Generate Genuine Scientific Creativity
- Rose Yu argues AI can create genuinely new scientific theories by distilling vast data and world knowledge into hypotheses.
- She cites LLMs generating novel math theorems, molecules, and hypotheses as evidence of creative potential.
Why GPUs Fuel Modern AI
- GPUs excel because deep learning is heavy on matrix multiplications, which GPUs accelerate by orders of magnitude.
- Rose Yu notes the same deep learning algorithms power LLMs, diffusion models, trajectories, and physics discoveries.
From L.A. Commute To Google Maps Deployment
- Rose Yu recounts building the Diffusion Convolutional Neural Network inspired by fluid diffusion to forecast traffic.
- The model extended accurate forecasts from 10–15 minutes to about one hour and was deployed by Google Maps.
