Liquid AI

Liquid AI, founded by MIT CSAIL researchers, is a Boston-based AI company specializing in creating a new generation of foundation models that surpass traditional Generative Pre-trained Transformers (GPTs). Their innovative approach integrates principles from biology, physics, neuroscience, mathematics, and computer science, leading to the development of their flagship product, Liquid Foundation Models (LFMs). These models are built on dynamical systems, numerical linear algebra, and signal processing principles, enabling them to handle various types of sequential data including text, audio, images, video, and signals efficiently.

Liquid AI's models feature a novel architecture that allows parameters to adapt and change over time through experience, making them highly adaptive. They also maintain a significantly smaller memory footprint compared to traditional large language models (LLMs), requiring less computational power and storage. Additionally, Liquid AI's models are capable of processing and understanding multi-modal data, making them versatile for various applications.

The company is currently in the process of launching its products, with plans to provide on-premises and private AI infrastructure for enterprise customers. They also encourage users to participate in red-teaming efforts to test and improve their models.