The eloquence and human touch in text responses generated by Chinese start-up DeepSeek’s artificial intelligence (AI) models, which have quickly gained popularity, can be partly credited to a small team of liberal arts graduates who curated the training data, according to former employees and analysts.
Advertisement
DeepSeek has sent shock waves through Wall Street and Silicon Valley and sparked a frenzy of AI adoption in China. While the Hangzhou-based company is known for offering generous compensation packages to attract talent in algorithms and computing, it has also assembled a small team of “data omniscients”.
Employees holding the peculiarly named role are tasked with sourcing knowledge in history, culture, literature and science to build a vast virtual library.
Wang Zihan, a former DeepSeek employee, said in a live-streamed webinar last month that the role was tailored for individuals with backgrounds in literature and social sciences. They provide insights on various data sets for model training, infusing a human touch into the company’s low-cost but high-performance models. “They contribute to efficiency and nurture ideas across all teams,” Wang said.
According to a recent social media post by another former DeepSeek employee, Zheng Size, some of the roles are filled by Chinese language and literature graduates, who have significantly improved the quality of Chinese content generated by DeepSeek through their carefully curated training data.
The first text generated by DeepSeek that went viral on the Chinese internet was a well-crafted post attributed to its founder, Liang Wenfeng, addressing claims that DeepSeek had changed China’s national fate.
Advertisement