Today, DereseEek is one of the leading AI companies in China that does not rely on funding from tech giants such as Baidu, Alibaba, or crowdfunding.

A young group of geniuses eager to prove themselves

According to Olang, when he targeted the Dreepsheek research team, he wasn’t looking for experienced engineers to build a consumer-facing product. Instead, he focused on PhD students from China’s top universities, including Peking University and TsingHua University, who were eager to prove themselves. Many have published in top journals and won awards at international academic conferences, but lack industry experience, according to the China Tech Publication Qbitai.

“Our tech positions are technically filled with people who graduated this year or in the last year or two,” Liang said 36Kr in 2023. The leasing strategy has helped create a company culture where people are free to use a lot of computing resources to pursue research projects. This is a different way of operating from established Chinese Internet companies, where teams are constantly competing for resources. (A recent example: A former intern was charged-A prestigious Awalicic Award Wonner, without sabotaging the work of his colleagues to prevent more computer resources for his team.)

Liang said students would be a better fit for high-investment, low-interest research. “Most people, when they’re young, can devote themselves to a mission without utilitarian considerations,” he explained. His pitch on future tricks is that Dereseeek was created to “solve the world’s toughest questions.”

The fact that these young researchers are almost entirely educated in China adds to their drive, experts say. “This young generation also embodies a sense of patriotism, especially as they navigate US restrictions and focus on technical hardware and software technologies,” Zhang explained. “Their determination to overcome these obstacles shows not only personal ambition but also a broader commitment to advancing China’s position as a global leader in innovation.”

Innovation born from a crisis

In October 2022, the US government began an export ban that prevented Chinese AI companies from accessing cutting-edge chips such as NVIDIA’s H100. The move presents a problem for Dereekeek. The company started with a stock of 10,000 H100’s, but it needs more to compete with companies like Openi and Meta. “The problem we are facing is not yet financing, but controlling the export of advanced chips,” Liang told 36kr In a second interview in 2024.

Deresteek needs to come up with a more efficient method to train its models. “They’re optimizing their architectural models using a battery of communication schemes – conventional methods of engagement,” said Wendy Change, a software engineer who runs the policy analyst at The Mercator Institute for in Chinese studies. “Many of these methods are not new ideas, but combining them successfully to create a cutting-edge model is a unique factor.”

Dereekek has also made important advances in multi-head attention (MLAt) and mixed experts that are more effective in improving computational efficiency. In fact, the latest Dereeseek model is so efficient that it needs one-tenth the computing power compared to the Meta 3.1 training model, According to Research Institute Epoch Ai.

DereSeek’s willingness to share these changes publicly has earned it a lot of goodwill in the AI ​​research world. For many Chinese AI companies, the development of Open Sure Surefers is the only way to play with their counterparts with users and contributors, who in turn help the models. “They are now showing that cutting edge models can be built with less, if not more, money and current optimization repair techniques,” Chang said. “We’re sure to see more testing in this direction going forward.”

The news could spell trouble for current US export controls targeting the manufacturing of computing resource bottlenecks. “There’s an estimate of how much computing power AI has, and what they can do with it, it can be great,” Chang said.