The Best New AI Products

Liquid AI, MIT Spinoff, Raises $37.5M to Advance Liquid Neural Networks

A recent MIT spinoff, Liquid AI, has raised $37.5 million to pioneer all-purpose artificial intelligence using liquid neural networks. Co-founded by robotics pioneer Daniela Rus, the startup values itself at $303 million post-money. Liquid neural networks, inspired by roundworm brains, offer smaller size and simpler construction, requiring less processing power. With fewer neurons and parameters, interpretability is enhanced, allowing for applications like drone navigation on devices as small as a Raspberry Pi.

Liquid AI's distinctive feature lies in the capacity of its liquid neural networks to modify parameters over time for better adaptation. Experiments have demonstrated superior performance in future value prediction, particularly in autonomous navigation scenarios. The startup aims to commercialize its liquid neural network architecture, competing with foundation model companies like OpenAI. The focus is on offering clients on-premises and private AI infrastructure, providing capital-efficient, comprehensible machine learning models for various applications.

The architecture's potential applications extend beyond drone deliveries to areas such as search and rescue, wildlife monitoring, weather patterns, financial transactions, medical readouts, and power grids. Liquid AI plans to grow its team and develop new liquid foundation models, leveraging the seed funding to push the boundaries of computational feasibility in 2022.

Headquartered in Boston with offices in Palo Alto, Liquid AI anticipates expanding its team to 20 members by the beginning of the next year. The startup envisions delivering more capable, dependable, and versatile AI models that prioritize safety and accountability in the evolving landscape of artificial intelligence.