Anandkumar Walks Away From Bezos. She Wants Physics, Not Chatbots. The Premise Is Controversial, Obviously.
Caltech professor Anima Anandkumar and Benedikt Jenik launched Accelerated Understanding Inc, a startup betting that AI's next leap comes from understanding the physical world, not from larger language models. They turned down an offer to join Jeff Bezos-backed Project Prometheus to pursue this independently. The company aims to build AI systems that learn from physical environments.
This illustrates the concept of grounding in AI. Language models predict tokens. They do not understand gravity, material properties, or spatial constraints. The mechanism Anandkumar proposes is embodied learning: training systems that interact with physical data rather than text corpora. The mental model here is that intelligence is not purely linguistic. Some problems require a body, or at least a physics engine.
Anima Anandkumar, Caltech professor and AI infrastructure engineer, and Benedikt Jenik founded Accelerated Understanding Inc. They declined an offer from Jeff Bezos-backed Project Prometheus.
- Go to Google's Teachable Machine website and select the Image Project option.
- Train a simple model by holding different household objects in front of your webcam and labeling each category.
- Test the model by introducing a new object under different lighting conditions. Expected outcome: you observe how a model trained on visual, physical data behaves differently from a text-only system, giving you a small taste of what non-linguistic AI learning looks like.