Anandkumar and Jenik launch Accelerated Understanding after turning down a 35% Prometheus stake
Anima Anandkumar and Benedikt Jenik unveiled Accelerated Understanding Inc, whose neural-operator model — no Transformer architecture — took in 5 trillion pieces of data in a single prompt during testing, roughly 5 million times what Anthropic's and Google's flagship models handle. The pair walked away from a late-2024 Project Prometheus offer worth a 35% stake and a $1 million salary rising to $2 million, and Bezos-backed Prometheus went on to raise a $12 billion Series B in June 2026. The bet is that one physics-native model can replace bespoke math models across chip design, robotics, extreme weather, and geological analysis — a different path to world models than the text-trained systems LeCun and Fei-Fei Li are building.
Source: tech.yahoo.com ↗
The language-centric view of intelligence is humans at the center. Putting physics at the center is a nature-centric view.
Why this matters
- → Physics-native AI model handles 5 million times more data than language models, enabling single systems to rep
- → Discards Transformer architecture for neural operators—a fundamental departure signaling that text-trained AI
- → Founders' decision to build independently over a $12 billion opportunity suggests confidence in physics-first