At a Glance
- Anima Anandkumar and Benedikt Jenik launched Accelerated Understanding Inc. instead of joining Jeff Bezos’s Project Prometheus
- Their AI processed 5 trillion pieces of data in one prompt, about 5 million times more than typical flagship models handle
- The system skips the Transformer architecture behind ChatGPT and runs on neural operators instead
- Bezos’s team had offered a 35% company stake and over $2 billion in funding
- Target markets include chip design, robotics, weather forecasting, and energy exploration
A pair of scientists once courted to lead one of Silicon Valley’s most heavily funded AI ventures chose to go their own way – and on Tuesday, they showed the world why.
Accelerated Understanding Inc., founded by Anima Anandkumar and Benedikt Jenik, unveiled an AI model built not for language, but for physics. In testing, it handled 5 trillion pieces of data in a single prompt, dwarfing what models from Anthropic and Google can typically process.
Key Details
Rather than predicting the next word in a sentence, the model predicts physical behavior across space and time. It’s built on neural operators, technology Anandkumar helped develop during her years at Nvidia, where CEO Jensen Huang personally encouraged her to pursue the idea.
Back in 2024, investor Vik Bajaj offered the pair a leading role in what became Prometheus, backed by Bezos and eventually raising $12 billion. They declined.
Why It Matters
Accelerated Understanding is pitching its model for chip design, robotics, and energy applications – sectors where physics-based AI could cut costly trial and error.
What to Watch
Whether enterprise clients bite, and how Prometheus responds now that a rival vision for “world models” has entered the race.
Stay ahead of every market-moving headline with QuoMarkets.
Source: Reuters
