- Liquid AI aims to bring its founders’ research on liquid neural networks—a novel kind of AI that uses a lot less power and can execute specific jobs more reliably than classic models—to the market.
- The reports state that Liquid AI plans to develop commercial foundation models with the funds from its recently disclosed investment round.
Liquid AI Inc. is creating artificial intelligence models using a design known as a liquid neural network. The startup recently revealed that it has raised USD 37.6 million in seed funding.
PagsGroup and OSS Capital spearheaded the investment. Along with them were Bob Young, the Co-founder of Red Hat, Automattic Inc., a WordPress developer, Samsung Electronics Co. Ltd.’s Next fund, and over six additional backers. The investment was reportedly finished at a USD 303 million valuation.
Chief Executive Officer Daniela Rus, an MIT Computer Science and Artificial Intelligence Laboratory Director, oversees Liquid AI. Together with three other researchers from the lab, Rus formed the business early this year. Liquid AI aims to bring its founders’ research on liquid neural networks—a novel kind of AI that uses a lot less power and can execute specific jobs more reliably than classic models—to the market.
Artificial neurons, comparatively simple code snippets, make up AI models. These little pieces of code carry out specific tasks allocated to the AI model they operate under. An equation, or a set of equations, that differs between neural networks governs how each neuron behaves.
The equations underlying the neurons in liquid neural networks, such as the ones Liquid AI is creating, can be changed. They can also alter the way those neurons communicate with one another. Liquid AI claims that these neural networks are more adaptive than conventional AI models since they can change their own design.
AI models are typically limited to the tasks they were designed and trained for. For instance, an autonomous driving system accustomed to working in pleasant weather would find it challenging to function dependably in the rain. Processing errors are less likely when using liquid neural networks since they are easier to adjust to changing circumstances.
Efficiency-wise, the technology also offers an advantage. Compared to typical AI models, liquid neural networks can be created with a substantially smaller number of neurons and require fewer parameters—configuration settings that dictate how input is handled. As a result, less infrastructure is required to run them.
The fact that liquid AI models have very few neurons also makes it easier for researchers to comprehend the interactions between those neurons. Consequently, deciphering the reasoning behind an AI model’s choice and determining its accuracy becomes simpler.
The reports state that Liquid AI plans to develop commercial foundation models with the funds from its recently disclosed investment round. It also intends to introduce a platform to let users create their own liquid neural networks. The company will add eight personnel in the following months to help with its go-to-market initiatives. Currently, it works with 12 employees on board.