New AI Learns Physics, Not Language

Accelerated Understanding Inc. has launched a novel AI model capable of processing an immense 5 trillion data points in one go, a figure millions of times larger than leading language models can handle. This groundbreaking system, co-founded by Caltech professor Anima Anandkumar and engineer Benedikt Jenik, distinguishes itself by focusing on physics phenomena rather than language comprehension.

Anandkumar, who previously worked at Amazon and Nvidia, drew inspiration from her work on neural operators. This physics-centric approach contrasts with language-based AI like ChatGPT, aiming to predict physical occurrences and provide nature-centric insights. The company believes this capability can revolutionize industries.

Potential applications include optimizing chip design by understanding material physics, advancing robotics, predicting extreme weather patterns, and analyzing geological data for energy firms. Accelerated Understanding is prioritizing enterprise clients over a consumer offering.

This development follows discussions with an investor involved in Jeff Bezos's Project Prometheus, which explored a potential collaboration. Anandkumar and Jenik ultimately chose to continue their work independently, while Prometheus secured significant funding for its own AI manufacturing goals.