The past one week has seen a Chinese AI startup, DeepSeek, take the artificial intelligence world by storm.
DeepSeek released its open-source AI model that is challenging industry leaders like OpenAI.
With just $6 million in funding, the company is making waves by outperforming established players in key areas such as math and reasoning benchmarks.
Founded by Liang Wenfeng, a 39-year-old computer science expert- this sudden rise highlights a twist in the ongoing tech cold war between the U.S. and China.
As American export restrictions tighten access to advanced GPUs like Nvidia’s H100, Chinese firms like DeepSeek are rewriting the rules of the game, proving that brute-force hardware isn’t the only path to innovation.
Liang Wenfeng’s education and early Life
Liang Wenfeng was born in 1985 in Zhanjiang, Guangdong, to a family deeply rooted in education—his father was a primary school teacher.
This early exposure to learning and curiosity shaped Liang’s academic pursuits. He attended Zhejiang University, one of China’s most prestigious institutions, where he earned a Bachelor of Engineering in electronic information engineering in 2007.
He continued his studies at the same university, completing a Master of Engineering in information and communication engineering in 2010. These formative years laid the foundation for his future endeavors in technology and innovation.
The beginning of DeepSeek
DeepSeek’s roots trace back to High-Flyer, one of China’s top quantitative hedge funds. Liang Wufeng, who holds a master’s degree in computer science, initially led High-Flyer’s deep-learning research branch, Fire-Flyer, which used supercomputers to analyze financial data.
In 2023, Liang decided to pivot Fire-Flyer’s resources toward building DeepSeek, aiming to develop artificial general intelligence (AGI). “I wouldn’t be able to find a commercial reason for founding DeepSeek,” Liang told Chinese tech publication 36Kr. “Basic science research has a very low return on investment. But we’re doing this because we believe in the mission.”
DeepSeek’s research team is composed primarily of recent graduates from top Chinese universities like Peking University and Tsinghua University. Many of these young researchers have already made a name for themselves in academic circles, publishing in top journals and winning international awards.
What to know
DeepSeek’s latest model, DeepSeek-R1, has surpassed OpenAI’s GPT-4 in several performance metrics, including capability, cost-efficiency, and openness.
- This achievement points to a shift in the global AI landscape, where Chinese firms, once seen as lagging behind due to U.S. export controls on advanced chips, are finding innovative ways to compete.
- Unlike many Chinese AI companies that focus on downstream applications, DeepSeek has taken a different approach. By optimizing its software and embracing open-source collaboration, the company has managed to create cutting-edge models without relying heavily on expensive hardware.
- Despite US restrictions, DeepSeek found ways to optimize its models using innovative techniques such as Multi-head Latent Attention (MLA) and Mixture-of-Experts.
- These methods have made DeepSeek’s models significantly more cost-effective, requiring just one-tenth the computing power of comparable models like Meta’s Llama 3.1.
Its decision to release its models as open source has earned it widespread goodwill in the global AI community. By sharing its innovations, the company is not only attracting users and contributors but also positioning itself as a leader in collaborative AI development.
This approach could pose a challenge to U.S. export controls, which aim to limit China’s access to advanced computing resources.
