Liquid AI, a startup emerging from MIT, has unveiled a series of new AI models that aim to redefine the capabilities of neural networks. The company’s approach is inspired by the nervous system of the C. elegans, a tiny roundworm known for its simple yet effective neural structure. According to the company the new technology is more efficient, less energy-intensive, and more transparent than traditional AI models.
A New Kind of Neural Network
At a recent event held at MIT, Liquid AI showcased its latest advancements in AI technology. The company has developed a novel type of “liquid” neural network, which operates differently from conventional models. In traditional neural networks, each simulated neuron has a static value or “weight” that influences its function. In contrast, liquid neural networks utilize equations that predict neuron behavior over time, allowing for a more dynamic and adaptable learning process.
Ramin Hasani, co-founder and CEO of Liquid AI, explained that this technology was once merely a research project. “It was once just a science project, but this technology is fully commercialized and fully ready to bring value for enterprises,” Hasani stated. The new models include applications for fraud detection in financial transactions, self-driving car control, and genetic data analysis.
Liquid AI’s approach allows its networks to learn continuously, even after the initial training phase. This flexibility is a significant advantage over traditional models, which typically require retraining to adapt to new data. Additionally, the liquid neural networks are designed to be more transparent, enabling users to trace back the decision-making process of the AI.
Performance and Future Potential
The company has made strides in performance as well. In September, Liquid AI revealed large language models based on its liquid network design. One of these models, with 40 billion parameters, outperformed a 70-billion-parameter model from Meta on a benchmark known as MMLU-Pro. Sébastien Bubeck, a researcher at OpenAI, remarked, “The benchmark results for their SLMs look very promising.”
Despite the excitement surrounding Liquid AI’s technology, there are some important caveats. The liquid networks are particularly suited for tasks involving temporal data, and adapting them for other types of data may require additional coding. Furthermore, convincing large corporations to adopt this new AI design poses another hurdle.
Hasani emphasized the importance of demonstrating the benefits of their technology, stating, “We are getting into stages where these models can alleviate a lot of the socio-technical challenges of AI systems.” The company is eager to show that the advantages of efficiency, transparency, and reduced energy costs outweigh the challenges.
A Unique Position in the AI Landscape
Liquid AI’s technology stands apart from that of major players like OpenAI and Google. The startup’s models are designed to operate with significantly less power, addressing growing concerns about the environmental impact of AI. As the demand for AI technology surges, so does the energy consumption associated with it. A study from last year projected that data centers powering AI could consume as much electricity as entire countries by 2027.
Governor Maura Healey, who attended the event, expressed her enthusiasm for the potential of Liquid AI. “It reminds me of the shots first fired in Concord and Lexington that gave birth to this great country,” she said to the cheering crowd. “We have been about revolution and innovation from the very beginning,” she added.
Liquid AI has raised nearly $50 million in funding so far but this is unlikely to be enough to win the AI race. Mikhail Parakhin, chief technology officer of Shopify, noted that securing the billions required for that purpose would be a significant challenge. “Getting that would be a tall order at this point,” he said, while acknowledging the remarkable technology Liquid AI has developed.
Real-World Applications
At the MIT event, Liquid AI showcased practical applications of its technology. One demonstration featured a chatbot that operated entirely on a smartphone without internet access, engaging users in discussions about AI-themed Halloween costumes. Another model quickly analyzed financial data to identify suspicious cryptocurrency trends.
The event attracted a diverse audience, including potential investors and entrepreneurs, all eager to witness the capabilities of Liquid AI’s technology. Among the people present was Steve Pagliuca, co-owner of the Boston Celtics and an investor in the company.
Yvonne Hao, Massachusetts secretary of economic development said that the stakes in this technological race are indeed very high. “AI is the next frontier we in Massachusetts will lead,” said Hao.
(Image by K. D. Schroeder)
