DeepSeek: An Engineer Confronts AI That Could Surpass Him

Key Takeaways

1Liu Shengyu, an engineer at DeepSeek, describes how his profession is being transformed by AI

2He expresses his technical pride but fears the loss of hands-on practice in favor of supervising agents

3He warns against the concentration of control over models and advocates for open access

4The debate on AI safety intensifies as Chinese labs make rapid progress

Why it matters : This testimony sheds light on the dilemmas faced by AI engineers in the face of automation and the concentration of technological power.

Liu Shengyu, a contributor to a key component of the latest DeepSeek model, published a lengthy text in Chinese where he describes the shift in his profession towards supervising AI agents. He expresses pride in DeepSeek v4.1, but also his belief that AI-generated operators will match, if not surpass, his own. His message comes as the race for models accelerates and security alerts multiply.

Warnings and the Model Race Intensify

The controversy surrounding the trajectory of AI has recently intensified. Jacob Coxon, a former researcher at Anthropic and OpenAI, announced that he was leaving Anthropic for AI safety reasons. In a message posted on X, he claimed that some companies are rushing towards a self-improving superintelligence and playing with lives. These statements have fueled a debate involving Dario Amodei, Sam Altman, and Elon Musk regarding the speed of the race towards increasingly powerful systems. Meanwhile, China is ramping up its efforts to become a major force in cutting-edge AI. Following DeepSeek, Moonshot AI and its Kimi K3 model have reignited attention on the rapid advancements of Chinese laboratories.

From the Joy of Coding to Agents: The Profession AI is Reshaping

Liu Shengyu expresses satisfaction with the success of DeepSeek v4.1, while noting that in about a year, AI has evolved from a tool for research assistance and debugging to an entity capable of interpreting GPU code and optimizing software operators. According to him, it is likely that AI-generated operators will reach an equivalent, if not superior, level to those he writes himself. He does not foresee unemployment, but fears losing the practical aspect of his job in favor of a conductor role for agents. Writing kernels remains an intense source of pleasure for him, with the excitement of inventing techniques and the pride of surpassing the official versions from hardware vendors. However, he compares the transition to moving from hand knitting to machine knitting, where the joy of the craft is overshadowed by the noise of machines. Saying goodbye to the era of hand-written kernels and human ingenuity-driven optimization feels particularly painful to him, as he describes having intense attachments, even crying when leaving a rented apartment after a year.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Building What Could Replace: An Embraced Paradox

Liu openly questions the sense of perfecting kernels while knowing that it may accelerate his replacement by the models they serve to train and execute. Nevertheless, he believes that holding back would be futile, as competitors would advance and replace him regardless. In this perspective, he states that he prefers to be the one who "revolutionizes himself" if the revolution is inevitable.

Liu Warns Against the Concentration of Advanced AI in Few Companies

Liu warns that cutting-edge AI concentrated in the hands of a few companies could increase inequalities by giving an enhanced advantage to those who have access. He advocates for open and affordable advanced intelligence for all. He claims not to trust Anthropic or OpenAI to guarantee this access and expresses a desire that Anthropic does not master the most advanced AI or AGI, going so far as to compare, albeit in an exaggerated manner, such a scenario to the hypothetical situation of Hitler having atomic bomb technology before the Allies.

A Public Testimony from Inside a Chinese Lab

Liu Shengyu has widely shared his experience, offering an unusual glimpse into the life of an engineer within a Chinese AI laboratory. The author, who contributed to a key component of the company's latest model, fears retaining his job while losing a significant part of the profession. Published in Chinese on his public WeChat account under the title "I Have No Choice But to Bury My Talent in the Past," he embraces the strangeness of helping to build an AI that could obsolete his skills. He confirmed being the author and did not wish to comment further. Finally, he urges his readers on the necessity of interacting with AI to protect their livelihoods, while anticipating the disappearance of what they prefer in their work.