The Next Scaling Law After LLMs: Caltech's Anima Anandkumar Makes Her Case at Supercomputing 2026
Reported by Tech Times1 source ↓
At SC26, Caltech’s Anima Anandkumar argued that neural operators could address physical-science problems that text-focused AI models struggle with by learning mathematical functions from physical data.
Key points
- Anandkumar proposed neural operators as an approach to AI for physical-science problems
- The models predict phenomena across space and time, including fluid dynamics and temperature gradients
- Her lab has applied the approach to quantum chemistry, carbon capture modeling and black hole simulation
Details
Tech Times reports that Anandkumar made the case during a Tuesday morning address at SC26 in Chicago. She argued that language-model architectures and the data used to train them have limits when applied to physical systems such as weather, fusion reactors and semiconductor stress testing.
Neural operators learn mathematical functions rather than text patterns, and are designed to predict physical phenomena across three-dimensional space and time. Anandkumar introduced the approach in 2020; the article says her lab has since applied it to quantum chemistry, carbon capture modeling and black hole simulation.
The article says Anandkumar’s argument also appears in a paper published in Daedalus, the journal of the American Academy of Arts and Sciences. She and AI infrastructure engineer Benedikt Jenik founded Accelerated Understanding, which launched with a debut model on August 25; the article reports that the model processed 5 trillion data points in one prompt during testing.
Written from coverage as of
Images
- Tech Times