Intelligence on the Edge: Liquid AI's Ramin Hasani on the Search for Device-Native Foundation Models

Liquid AI optimizes artificial intelligence for extreme efficiency by leveraging biologically inspired, differential equation-based neural networks that prioritize adaptability over massive parameter counts. CEO and co-founder Ramin Hasani explains how these models, which originated from studying the nervous system of *C. elegans*, utilize input-dependent dynamics to achieve robust out-of-distribution generalization. The company employs an Automated Foundation Model Design process that removes human architectural bias, using hardware-in-the-loop testing to tailor models for specific edge devices like smartphones and automotive systems. By focusing on this massive, underutilized global compute market, Liquid AI delivers high-performance inference on limited hardware, as evidenced by successful deployments with partners like Shopify and Mercedes-Benz. This approach challenges the reliance on cloud-based frontier models, demonstrating that sophisticated intelligence can be achieved with significantly lower energy and computational footprints.