China’s AI ecosystem is being revolutionized by open-source technology, which is not only driving innovation but also fostering collaboration and showcasing the transformative power of shared knowledge, said an expert.
Gao Wen, director of Pengcheng Laboratory (PCL), a leading research institution in the southern Chinese city of Shenzhen, highlighted how open-source culture promotes collaboration between research institutions and major tech firms, benefiting both developers and users alike.
"After I make the platform open-source, others, who use it, have to share their resources. Once they do, it will make my original system even better. Then, on top of all the open-source stuff, we will keep iterating. Many hands make light work. Everyone involved is adding fuel and stoking the fire," said Gao.
He noted that the firm is leveraging the "Pengcheng Cloud Brain" intelligent computing platform to train the PCL Mind Series of Large Models using an open-source approach, aimed at establishing it as a foundational application for China's computing power network.
The Pengcheng Cloud Brain II computing platform that has cultivated Pengcheng Mind also embraces an open concept. Specifically, 50 percent of its computing power is dedicated to self-developed projects within the Pengcheng Laboratory. Meanwhile, 40 percent is made available to domestic partners, research institutions, and universities, while the remaining 10 percent is allocated to the Shenzhen Municipal Bureau of Industry and Information Technology for use by external developers, Gao explained.
He emphasized the vital role of adopting an open-source approach in advancing software development.
"I advocated for open-source software about 20 years ago. At that time, within the national '863' program (National High-tech R and D Program), I was a firm proponent of opening source code. In my understanding, opening source might be more in line with the direction of human social development in terms of software ecosystem iteration, application promotion, and so on. This is similar to knowledge. In the past, many things were passed down from generation to generation among artisans and craftsmen. If any generation failed to pass it on properly, it would be lost. But now, the knowledge we learn is accessible to everyone, not just a few people. This is because knowledge is open-source. In fact, software is the same. If software becomes like knowledge, it should be open-source. Only by being open-source can it be iterated. Of course, it doesn't mean that once you open source it, you have no business opportunities and will fail. That's not the case. Teaching is also a means of making a living. Knowledge itself is open source, but teachers can still survive. It's the same with software. If software wants to develop rapidly in the future, it must be open-source," he explained.
While the "Pengcheng Cloud Brain II" is operating smoothly, Gao has already begun to focus on new challenges. Now, he and his team are in collaboration with relevant enterprises to develop the next-generation super-intelligent computer "Pengcheng Cloud Brain III," which has an intelligent computing power target of 16,000 petaflops per second, 16 times more powerful than "Pengcheng Cloud Brain II".
"It can be used for the training and application of large multimodal models. Currently, the intelligence of large language models is only language intelligence, while other forms of intelligence are relatively weak. In life, there are many things that people cannot express through language. That's why for many things we learn, we have to see them in person, experience them, and explore them. We usually say that things perceived through hearing, vision, or physically are difficult to describe clearly through language. Therefore, besides language, a multimodal model is one that combines these other elements with language for training. In this way, the overall intelligence level will gradually approach that of humans. We also hope that through the construction of this machine, it can be helpful to the domestic ecosystem," he noted.
Open-source AI to transform China's innovation landscape: expert
