The Chinese finance whizz whose DeepSeek AI model stunned the world (www.nature.com)

In January, China-based DeepSeek launched its groundbreaking R1 model, a reasoning large language model (LLM) that has sparked intrigue within the AI community. Founded by Liang Wenfeng, a former financial analyst turned entrepreneur, DeepSeek has quickly demonstrated that the U.S. may not hold as significant an edge in AI development as previously thought. The R1 model excels at complex problem-solving in areas like math and coding, thanks to its step-by-step reasoning approach, and is notably available as open weight, allowing researchers to download and modify it for their own projects. The release of R1 comes at a time when China's ambitions in LLM development faced hurdles, particularly with U.S. export controls on essential AI training hardware. However, Liang's foresight in acquiring 10,000 NVIDIA GPUs positioned DeepSeek to thrive. Moreover, the model's transparency is further exemplified by its peer-reviewed publication of training methodologies, which fosters collaboration and innovation in AI research. As DeepSeek's influence grows, this launch not only highlights significant advancements in model efficiency and cost-effectiveness but also encourages a shift toward open-source contributions, challenging traditional norms in the competitive landscape of AI development.