How Chinese AI Startup DeepSeek Made a Model that Rivals... | WIRED
A Chinese hedge fund's AI lab beats Western giants using fewer chips.
Deep Dive
DeepSeek, a lab spun out of Chinese quant hedge fund High-Flyer, released open-source model DeepSeek-R1 on January 20. It outperforms OpenAI o1 on several math and reasoning benchmarks. Founder Liang Wenfeng prioritized efficient software optimization over scaling, hiring young PhD graduates from top Chinese universities like Peking and Tsinghua who lacked industry experience. Backed by High-Flyer, DeepSeek shows that resource-constrained innovation can compete despite US export restrictions.
Key Points
- DeepSeek-R1 beats OpenAI o1 on math and reasoning benchmarks, despite US chip export restrictions.
- The model is open-source and built by a team of young PhDs from Peking and Tsinghua universities.
- DeepSeek is funded by High-Flyer, a top Chinese quant hedge fund with $8B assets; founder Liang Wenfeng prioritizes science over profit.
Why It Matters
Shows that efficient software innovation can overcome hardware bans, reshaping the global AI race.