Under the flood of AI, even DeepSeek's top engineers are worried about their future careers. DeepSeek machine learning system engineer Liu Shenghe published the long article “I Had to Bury My Talent Yesterday”. In the article, he mentioned that in the face of rapid progress in AI capabilities, he will not “lose his job” in the future, but he may have to “switch jobs.” This anxiety comes from an engineer at the cutting edge of the AI wave. Judging from his resume, Liu Shenghe has been designing and optimizing machine learning systems and operators at DeepSeek since April 2025. Prior to that, he was a Peking University 2021 Turing class student, was the captain of Peking University's supercomputing team, and led the team to the top two places in the world-class college supercomputer competition. DeepSeek released a new v4.1 model a few days ago, taking its capabilities one step further. Liu Shengyou mentioned that the main attention operator of DeepSeek v4.1 was written by himself, and he is proud of the success of DeepSeek v4.1. But what he knows is that in the first half of the year or a year, the operators written by AI are likely to be as good as they are, or even surpass themselves.

Under the flood of AI, even DeepSeek's top engineers are worried about their future careers. DeepSeek machine learning system engineer Liu Shenghe published the long article “I Had to Bury My Talent Yesterday”. In the article, he mentioned that in the face of rapid progress in AI capabilities, he will not “lose his job” in the future, but he may have to “switch jobs.” This anxiety comes from an engineer at the cutting edge of the AI wave. Judging from his resume, Liu Shenghe has been designing and optimizing machine learning systems and operators at DeepSeek since April 2025. Prior to that, he was a Peking University 2021 Turing class student, was the captain of Peking University's supercomputing team, and led the team to the top two places in the world-class college supercomputer competition. DeepSeek released a new v4.1 model a few days ago, taking its capabilities one step further. Liu Shengyou mentioned that the main attention operator of DeepSeek v4.1 was written by himself, and he is proud of the success of DeepSeek v4.1. But what he knows is that in the first half of the year or a year, the operators written by AI are likely to be as good as they are, or even surpass themselves.