Liquid AI: MIT's $2B LLM Startup | Agentic EP 10

Liquid AI develops foundational models using liquid neural networks, a technology inspired by the biological neurons of small species like the *C. elegans* worm. By utilizing differential equations rather than traditional transformer architectures, these models achieve high performance with significantly fewer neurons, enabling efficient, private, and low-latency AI deployment on edge devices such as satellites, autonomous vehicles, and mobile hardware. Unlike cloud-dependent models that incur massive infrastructure costs, this approach allows for local intelligence that maintains data privacy and operates in disconnected environments. Co-founders Ramin Hasani, Alexander Amini, and Daniela Rus emphasize that this shift toward "physical intelligence" addresses critical gaps in current AI, particularly in safety-critical sectors like finance and national security. By prioritizing efficiency and first-principles mathematical foundations, the company aims to commoditize intelligence at the edge, moving beyond the limitations of large-scale, centralized generative AI systems.