“DeepSeek”: How a Small Chinese AI Company Shocked Silicon Valley

In a surprising turn of events, a small Chinese startup has stunned AI developers in Silicon Valley by launching an advanced version of their inference model, a relatively new area of AI research.

“DeepSeek” released its model, dubbed “R1,” detailing in a comprehensive paper how to build a large language model on a limited budget that can learn and improve itself autonomously without human supervision.

American companies, including “OpenAI” and Google’s “DeepMind,” have been leading the development of inference models, a field of AI research aiming to align models with human cognitive abilities.

Founded by hedge fund manager Liang Wenfeng, DeepSeek unveiled its R1 model on Monday, outlining in detail how to construct a large language model affordably and capable of self-improvement without human oversight.

DeepSeek’s R1 release has sparked intense debate in Silicon Valley about whether better-resourced American AI companies, including Meta and Anthropic, can maintain their technical advantage.

Meanwhile, Liang has become a focal point of national pride at home. This week, he was the only AI leader invited to a high-profile meeting with the country’s second-most powerful leader, Li Qiang. Business leaders were urged to “focus their efforts on breaking through key core technologies.