DeepSeek is not an idealistic company, it just serves reality to ideals

DeepSeek’s answer is not to just look up at the stars or keep your feet on the ground, but to use practical problems to support a goal that is far enough away. Written by Li Jiaxing Liu Chenyi Editor | Zhang Wei Head image source | DeepSeek Cover source | DeepSeek DeepSeek is once again pushed into the spotlight. This company was once almost the most mysterious existence in the AI ​​​​circle: it has an engineer culture and a low profile. The founder rarely makes public appearances, and the team does not tell stories or do communication as frequently as typical Internet companies. But in the past year, it has become the center of the Chinese technology circle almost every once in a while. When the model was released, it was popular because of its low cost and open source; when the financing news came out, it sparked discussions because of its valuation and investors; this time, the recording of Liang Wenfeng’s nearly four-hour investor exchange meeting leaked, causing the outside world to re-examine the company. This time, what the outside world saw was a more complete and more contradictory DeepSeek. One side is a far enough target. Liang Wenfeng talked about AGI, about vision, about "goodwill towards the world", about not pursuing profit maximization, and about the inability to organize people without open source. He said that DeepSeek does not want to be the next super App, nor does it want to be the next Byte or Tencent; the company operates on vision rather than KPI, and researchers have a lot of free time to explore and do not work overtime. The other side is down to earth. Liang Wenfeng also talked about API pricing with a ten-month payback, how financing can alleviate the risk of team stability, how money must be turned into a card as soon as possible, the production bottleneck of domestic computing power, half of the core researchers bidding on data, and the organizational structure that must be supplemented as the company grows larger. He even said that in the worst case, selling API can support a listed company. How can an AI company look up to the stars and keep its feet on the ground? This is exactly the answer DeepSeek wants to give. According to the "Transcript of Liang Wenfeng Investor Exchange Meeting" leaked on the Internet, "Future Human Laboratory" made the following summary. Look far away, but break it down into steps. The long-term goal is AGI, but it’s not about talking about the distance in vain, but it’s broken down into continuous steps. Liang Wenfeng defined the long-term goal as AGI, but he didn’t define AG.