Liquid AI Develops Transformer Replacement Technology With Energy Saving Solutions

JAKARTA Artificial Intelligence Technology (AI) is getting more sophisticated today, but its energy consumption is still a challenge. According to Liquid AI, a MIT-fraction company, this high energy consumption needs to be overcome.

Today, almost all AIs are generatively dominated by the Transformer architecture. This architecture controls 99 percent of the world's AI systems, but this technology has one big weakness, namely weak efficiency due to high energy consumption.

Liquid CEO AI Ramin Hasani considers that Transformers cannot be a sustainable long-term foundation. The reason is, the way Transformers process data makes AI systems expensive and operationally inefficient.

"The problem is, Transformers process data quadratically. The more information is processed, the energy consumption increases exponentially," Hasani told VOI on the sidelines of the Expand North Star 2025 event in Dubai.

To answer these challenges, Hasani and his team at the Massachusetts Institute of Technology (MIT) developed a technology called liquid neural networks. This technology is inspired by the biological systems of humans and animals.

This system is able to develop intelligence and adapt to minimal energy consumption. According to Hasani, the way this technology works is completely similar to the way humans and animals do learning.

"If you look at humans or animals, they can learn and adapt very efficiently. They don't need great energy to control behavior or think. We imitate that biological principle," Hasani explained.

Liquid AI is now trying to pioneer the use of liquid neural networks for various applications. They apply this technology into smart cars to satellite devices. This technology can run efficiently on small devices without the need for access to a data center or cloud.

We built a different Liquid Foundation Model with GPT. Our model can be run on almost all devices, from Raspberry Pi, coffee machines, cars, cellphones, laptops, to satellites, "said Hasani. You can imagine an offline version of ChatGPT that can work without internet, directly on your phone.

During the development of this new technology, the Focus of Liquid AI is not only on energy efficiency. They also focus on other values, one of which is economic efficiency. This efficiency needs to drastically reduce AI's operational costs.

Next, Liquid AI also prioritizes data privacy, especially since their AI can run directly on the user's device. The technology they develop does not send personal data to external servers.

Lastly, Liquid AI is trying to make its technology more instant to use, but with high security. Hasani also stressed that AI on devices should also focus on low latency.

In addition to efficiency, Hasani emphasized the importance of transparency and interpretation in AI. He assessed that the biggest danger of AI is the lack of human understanding of the way AI makes decisions.

"We want every AI decision to be explained. When humans don't understand how AI thinks, that's where the danger arises," Hasani said. Our mission is to ensure AI remains in human control and is used wisely.