Liang Wenfeng’s blueprint for DeepSeek’s AI revolution

An innovator, not a follower. That is undoubtedly the best way to describe the approach of Liang Wenfeng, 40, founder of DeepSeek.

Par

Liang Wenfeng, le fondateur de DeepSeek, sur une image diffusée sur la Télévision centrale de Chine, le 20 janvier 2025. Crédit: CCTV

Originating from Zhangjian, a mid-sized city in Guangdong province, Liang Wenfeng founded a high-frequency trading company, High-Flyer, at just 23 years old, laying the groundwork for his future experiments in the field of artificial intelligence (AI).

Low-key – very few photos and public statements about him circulate – this trained engineer found the formula for success through an original managerial approach in the AI sector. Unlike his competitors, including Chinese ones, Liang Wenfeng primarily bets on local talent and avoids hiring AI veterans.

à lire aussi

Recent graduates from top universities, PhD students in their fourth or fifth year, and early-career professionals, just a few years out of school,” he explained to the Chinese media outlet 36Kr. “DeepSeek does not hire big names, as they rarely have the drive to innovate,” added a company recruiter in the same article.

Only the elite

Liang Wenfeng selects only the top talent from Chinese universities. And to attract them, he offers salaries among the highest in the industry

Young graduates, or even students, yes—but not just any: Liang Wenfeng selects only the elite from Chinese universities. And to attract them, he offers salaries among the highest in the market, sometimes exceeding those offered by the most generous employers in China’s tech sector, such as Bytedance (owner of TikTok, among others).

But that’s not all. Since its creation in 2023, DeepSeek has maintained a compact team of around 150 members, prioritizing collaboration in resource allocation and research project selection. The company operates under a model akin to an academic institution, where employees work on projects without fixed roles or rigid hierarchies.

Result: Instead of merely replicating existing models like ChatGPT, with a focus on strict performance metrics, DeepSeek has from the outset questioned the very foundations of AI models. This bold and efficient approach has led to the meteoric rise now associated with DeepSeek.

Written in French by Zakaria Choukrallah, edited in English by Eric Nielson

à lire aussi