DeepSeek founder Liang Wenfeng

1. Studied machine vision at Zhejiang University during undergraduate and graduate studies
2. In 2015, at age 30, founded the "High-Flyer" quantitative hedge fund
3. Accumulated considerable wealth through quantitative hedge fund trading (currently managing about USD 8 billion)
4. Wanted to build "human-level" artificial intelligence outside his main business, but was initially questioned when he proposed the idea to partners
5. In 2021, purchased 10,000 H800 chips and brought the top employees from his hedge fund to the new project team (these employees have extensive experience in maximizing Nvidia GPU performance)
6. In 2023, officially established DeepSeek, recruiting dozens of PhDs from top universities such as Peking University, Tsinghua University, and Beihang University. In China, offers top-tier compensation on par with ByteDance for technical talent, aiming to make DeepSeek a leading "local" AI company
7. Due to U.S. export controls, the DeepSeek team had to find alternative paths, using innovative training methods to make their LLM models (V3, r1) comparable in performance to OpenAI, Anthropic, Gemini, Grok, LLaMA, etc., at roughly one-twentieth of the cost
8. Although the training cost comparison is not entirely apples-to-apples, they significantly improved efficiency through original methods (there are also rumors of possibly borrowing from other models, having a larger H-100 cluster, or even receiving government support)
9. DeepSeek open-sourced and published their methods (the r1 reasoning paper lists over 200 authors)
10. DeepSeek recently topped the App Store rankings