My view on DeepSeek (2/N): Talent Strategy Early on, I met DeepSeek researchers focused on AIGC, including contributors to projects like DeepSeek Janus-series (Nov 2024 - Jan 2025) and DreamCraft3D #ICLR2024, and with one expert who helped optimize my paper (@xingchaoliu), he is also the author of rectified flow, which is been seen as the foundation of Stable Diffusion 3.0/3.5 @StabilityAI and Flux @bfl_ml. To my suprise, most researchers in DeepSeek were young—current PhD students or graduates within three years—hailing from Beijing’s academic elite with 3-5 top-tier publications. When asked why Liang Wenfeng (CEO of DeepSeek) prioritized youth, a DeepSeek colleague shared Liang’s philosophy: The enigma surrounding DeepSeek has sparked curiosity: What is its secret weapon? Foreign media suggest it's "young geniuses" – talents capable of competing with deep-pocketed American tech giants. In the AI industry, hiring seasoned veterans is the norm. Most Chinese AI startups prefer experienced researchers or those with overseas PhDs. DeepSeek, however, defies this trend by prioritizing young talent with minimal work experience. A headhunter who collaborated with DeepSeek revealed: "They avoid senior engineers – 3-5 years of experience is already the upper limit. Candidates with over 8 years' experience get rejected outright." In a May 2023 interview with 36Kr, CEO Liang Wenfeng confirmed: "Most of our developers are either fresh graduates or newcomers to AI. Core technical positions are predominantly held by recent graduates or those with just 1-2 years of experience." How does DeepSeek evaluate candidates without traditional credentials? The answer lies in potential. Liang Wenfeng once stated: "For long-term endeavors, experience matters less than foundational skills, creativity, and passion. While the world's top 50 AI talents might not yet be in China, we're committed to cultivating our own." This mirrors OpenAI’s early strategy. In 2015, Sam Altman recruited ambitious young researchers like Andrej Karpathy (Stanford PhD), John Schulman (Berkeley), and Wojciech Zaremba (NYU). These “young wolves” birthed breakthroughs like GPT (Alec Radford), DALL-E (Aditya Ramesh), and GPT-4o’s multimodal lead Prafulla Dhariwal. From left to right: Ilya Sutskever (former Chief Scientist), Greg Brockman (former President), Andrej Karpathy (former Technical Lead), Durk Kingma (former Researcher), John Schulman (former Head of the Reinforcement Learning Team), and Wojciech Zaremba (current Technical Lead). This “wolf pup strategy” has already paid off for OpenAI, giving rise to key figures such as “GPT Father” Alec Radford (a graduate of a third-tier private college), “DALL·E Father” Aditya Ramesh (an undergrad from NYU), and the GPT-4 multimodal lead, three-time Olympiad gold medalist Prafulla Dhariwal. Thanks to these young mavericks pushing boundaries, OpenAI—initially a relative unknown next to DeepMind—has carved out a path to survive and grow into a powerhouse, even when its original world-saving plan wasn’t entirely clear. Liang Wenfeng recognized Sam Altman’s successful strategy and decided to follow a similar path. Unlike OpenAI, which waited seven years to see ChatGPT, Liang’s efforts paid off in just over two years—a testament to “Chinese speed.”