The family office of Boston Celtics co-owner Steve Pagliuca is one of the lead investors in an MIT spinoff developing a new generation of AI foundation models.
The startup, Liquid AI, announced a $37.5 million investment in seed capital. The funding was led by Pagliuca's family office, PagsGroup, and OSS Capital, with additional investors such as Tobias Lütke, Shopify co-founder; David Blundin of Link Ventures; Breyer Capital and Tom Preston-Werner, Github co-founder.
“There’s a lot of people going at (AI), there’s a lot of hype, and there’ll be a few that succeed,” Pagliuca told BostInno. “I think Liquid AI has a really great shot at being a very disruptive AI technology that will be more cost effective than the current versions that are out there.”
Liquid AI was founded by Ramin Hasani, Mathias Lechner, Alexander Amini and Daniela Rus, AI and machine learning scientists from the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).
Amini and Rus had co-founded another AI startup, for detecting bias in AI and help make it more trustworthy, with fellow MIT researcher Elaheh Ahmadi in 2021. Themis AI announced a $2 million early-stage round led by E14 Fund and Mozilla Ventures.
Hasani, the Liquid AI’s chief executive, told BostInno the company is headquartered in Brookline and has 14 employees. Less than half of the team is in California, and the rest are in Massachusetts.
Hasani said he has been working on designing AI systems for years at MIT and while completing his Ph.D. program at Technische Universität Wien in Austria. Hasani and his colleagues have been looking at how to bring insights from brains and neuroscience into machine learning and the design of AI systems.
The result was the invention of liquid neural networks. Hasani said liquid refers to flexibility, which means the systems are able to stay adaptable even after they’re trained and are very capable even at a small size, which has speed and cost benefits for customers.
Hasani said they’re focused on “extending these ideas of flexible intelligent systems that are more efficient compared to the machine learning system that exist today.”
Liquid AI will focus on serving enterprises with its liquid foundation model. Hasani said this model is a general intelligence system. In some ways it is similar to ChatGPT which started with text, but Liquid AI’s model will be able to process text, images, videos, signals and other inputted information.
The company’s models will be able to solve tasks in different domains that go beyond the data they have seen during training, Hasani said. The chief executive said the company’s models are also more reliable and have greater explainability for how they arrive at their output than current machine learning models.
“What we want to offer in this space is a private, on premise infrastructure in which you can build liquid foundation models for your use cases for your enterprise,” Hasani said.
Sign up for The Beat, BostInno’s free daily innovation newsletter from BostInno reporter Hannah Green. See past examples here.
