中国的深度求索(DeepSeek)正以迅雷不及掩耳之势席卷全球。以下是推动这一人工智能热潮的幕后智囊团队。

Wed, 19 Feb 2025 01:08:00 GMT

2025年1月28日,DeepSeek位于北京的办公室。Peter Catterall | Afp | Getty Images

人工智能初创公司DeepSeek迅速崛起,震撼了全球AI界,但其团队在中国以外却鲜为人知。DeepSeek的创始人梁文峰被一些西方媒体称为中国的Sam Altman。但与硅谷的同行不同,梁文峰一直保持低调。梁文峰的团队由国内顶尖大学的年轻毕业生组成,同样鲜为人知。据中国官方媒体报道,该团队不到140人,尽管其最新R1推理模型的研究论文列出了约200名贡献者。CNBC无法确认团队的官方规模。据熟悉该公司的人士透露,除了核心技术人员外,DeepSeek大部分共享了母公司High-Flyer的高级管理团队、运营人员、人力资源部门和财务人员。

以下是这位AI轰动背后的人物以及这家初创公司如何诞生的概述。

梁文峰

随着DeepSeek的聊天机器人登上全球应用排行榜榜首,梁文峰最近几周获得了媒体的广泛关注。上个月,据报道他在家乡中国受到了英雄般的欢迎,并出现在中国总理李强主持的圆桌会议上,最近还出现在本周早些时候由习近平主席主持的闭门研讨会上。这位40岁的DeepSeek创始人对媒体相当回避,除了去年7月和2023年两次罕见地接受中国媒体36Kr的采访。这些采访描绘了一位理想主义领导者的形象,他致力于实现人工通用智能(AGI)——一种模仿人类能力的AI——并将中国转变为技术创新者。

梁文峰1985年出生,在中国南方的港口城市和贸易中心湛江长大。据当地媒体报道,他是一名全优学生,尤其在数学方面天赋异禀。梁文峰在初中自学微积分后,于2002年考入浙江大学,并于2010年获得信息与通信工程学士和硕士学位。

据曾采访梁文峰的中国科技媒体36Kr报道,梁文峰专注于机器视觉研究,2008年开始编写机器学习算法,分析市场趋势和宏观数据以做出投资决策。当时AI并不是典型的量化策略,但梁文峰从量化投资的先驱Jim Simons那里获得了灵感,Simons创立了全球最成功的基金之一文艺复兴科技公司,梁文峰在Simons传记的中文版序言中说道。

High-Flyer基金经理

2015年,梁文峰和大学好友金旭创立了High-Flyer资产管理公司,这是一家量化对冲基金,使用复杂的数学算法预测市场趋势并做出投资决策。金旭毕业于浙江大学竺可桢荣誉学院,该学院从这所精英大学中选拔顶尖学生。

金旭在博士期间专注于机器人自主导航和机器学习——与梁文峰的博士后研究方向相似——并且是中国探月计划视觉导航研究项目的关键成员。金旭曾在2010年代初在华为技术公司从事软件开发,现在领导High-Flyer的技术开发并制定交易策略,他在私募股权数据库PaiPaiWang上的个人资料页面显示。

High-Flyer首席执行官郑哲毕业于与梁文峰和金旭相同的大学,之后在伦敦经济学院获得硕士学位。 d Politics. Prior to High-Flyer, Lu worked at the state-backed China Merchants Bank, where he was engaged with macro research and overseas derivative investment.In an interview with Chinese state media in 2023, Lu said: “We have set up a new team independent of investment, what is equivalent to a second start-up” — which later grew to become DeepSeek. “We want to do things with greater value and things that go beyond investment industry.”The pair manage some of the best performing funds under the company’s portfolios, with averaged returns over 20% in 2024, according to PaiPaiWang. That was above gains of about 15% in the CSI 300 index last year, a 5% rise in the small-cap CSI 500. The quant fund’s profits were partially channeled to fund the rise of DeepSeek, Liang told 36 Kr in 2023.Brains behind DeepSeek In 2023, High-Flyer spun off DeepSeek as an independent enterprise, expanding its remit beyond investment and focusing on pursuing AGI. The team consists mostly of local engineering, computer science and AI graduates from top universities in China — such as Tsinghua University and Peking University — many of whom have published recent papers on subjects such as language models and machine learning.A number of team members are also graduates from top American universities with experience at Nvidia and Microsoft who decided to return to China’s growing AI industry, according to their LinkedIn profiles. A key attribute that sets the team apart is age, as DeepSeek favors graduates with less work experience.Instead, “they emphasize academic degrees, awards at international programming competitions, research papers published at top industry journals,” a headhunter for DeepSeek told CNBC.In the interview in 2023, Liang said experience is less important in the long run and “foundational abilities, creativity, and passion are more crucial.” In 2024, he said that while the top 50 talent in AI may not have been in China, DeepSeek was aiming to cultivate its own.Top graduates also appear to be attracted to the firm because of its reportedly higher salaries and greater degree of bottom-up management than what might be found at a larger tech firm. watch nowVIDEO7:0007:00DeepSeek was a ‘masterful gambit’ in AI brinkmanship between U.S. and China, saysStreet Signs Europe

原文链接:https://www.cnbc.com/2025/02/19/china-deepseek-origin-story.html

jpichiban