A DeepSeek engineer has offered a rare and deeply personal look at what it feels like to build increasingly powerful AI while knowing that the same technology could eventually make parts of his own job obsolete. In a sentimental letter published on his public WeChat account, China’s OpenAI and Anthropic rival – DeepSeek AI developer Liu Shengyu reflected on how quickly AI has changed his work. He said that within about a year, AI moved from helping him search documents and find bugs to reading GPU code and optimising software operators. Liu, who helped write a key component of DeepSeek’s latest model, said he is proud of DeepSeek v4.1 but also knows that AI-written operators could match or even surpass his work.His biggest question is why he continues to push the technology when faster and better AI could ultimately replace the work he loves. “Human beings have never hesitated when it comes to engineering their own destruction,” Liu wrote, before explaining why he would rather “revolutionize myself” than be left behind. His words offer a human view of the growing AI race and the fear that the technology could transform not just jobs, but the work people deeply enjoy.
Here’s the letter DeepSeek engineer shared
Below is the text of full letter DeepSeek engineer shared (translated from Chinese)“A few days ago, DeepSeek v4.1 was released, taking the capabilities of small models to a whole new level.The speed of AI development has far exceeded everyone’s expectations. From the initial version of ChatGPT, which could only chat in a babbling manner with a context length of only a few thousand tokens, to OpenAI o1, DeepSeek R1, and Kimi K1.5 Thinking with reasoning capabilities, only two years have passed; from reasoning models to intelligent agents that can now smoothly execute commands and complete complex tasks in various harness tools, only a year and a half has gone by. It’s hard to imagine what AI will be like in another year, two years, or three years, how powerful it will be, whether it will have already possessed the ability to self-evolve, and whether it will have deeply penetrated fields such as embodied intelligence.AI is getting better and better at writing operators.AI has also made rapid progress in my field of operator design and programming. In just one year, it has transformed from a simple assistant that could only help me look up documentation, read code, and find bugs into an operator master capable of independently reading CUDA, PTX, and SASS code, analyzing the pause time of each instruction using professional tools, and then independently optimizing operators. I believe that in the near future, it will also have the ability to independently design operator schedulers, evaluate the performance of different scheduling schemes, implement them, and optimize them.Of course I am proud of the success of DeepSeek v4.1—after all, I wrote its main Attention operator[1], and its excellence is an affirmation of my operator. However, the wheels of time keep rolling forward, and no one can stop the development of technology. I am well aware that in another six months or a year, the operators written by AI will most likely be as good as mine, or even surpass mine. AI can think of 300 tokens in one second, type out a line of commands in half a second, and write a piece of code in twenty seconds, while I cannot; AI can continuously improve in terms of model depth, thinking intensity, tool call volume (frequency of interaction with the environment), and even parallelism, while I cannot.Humans have always been ruthless in their self-destruction. Why, knowing that “the better my operators are, the faster our new models will train and infer, the faster their capabilities will improve, and the sooner I’ll be replaced,” do I still choose to optimize my operators? On one hand, writing operators is indeed like playing a video game, providing immense pleasure. The excitement I feel when inventing a new technology or seeing my operator’s performance improve is comparable to a speedrunner breaking their own record. At the same time, seeing my operator’s performance far surpass the manufacturer’s official operators fills me with immense pride. But beyond that, a more important reason is that even if I were to “give up” or deliberately sabotage model training, other models would continue to develop and ultimately destroy me. “Of course, I hope I won’t be revolutionized, but if I have to be, I hope I’m the one who revolutionizes myself.” With everyone so obsessed with self-destruction, I’m forced to join this brutal arms race.What about me?What will happen to me when AI’s ability to write operators truly surpasses mine?My assessment is that I won’t necessarily lose my job, but I’ll have to change careers. I can still keep my current job, but this might mean I’ll never have the chance to do the work I once loved again.I once made a judgment about the changes of the times and my personal future situation: because the times are changing so rapidly (the development of AI mentioned above is a good example), I have absolutely no way of predicting what will happen in five or ten years. But in any case, I believe that with my vision, judgment, initiative, and intelligence, I can stay at the table of the times and stand at the forefront of the times again. However, this judgment can only guarantee that I will not be “unemployed,” not that I will not need to “change careers.” Rather, this judgment encourages me to avoid unemployment by changing careers.What does this career change mean? It means I need to abandon my long-cultivated and passionate field of operator design, writing, and optimization, and instead become an “Agent pilot.” Previously, my interests, my strengths, and the industry’s needs were largely aligned; now, AI has made what I’m good at even better, and the industry’s demands have shifted from “people who can write high-performance operators” to “people who can use AI to produce high-performance operators faster.” To adapt to the industry’s needs, I must abandon my previous passion and turn to an unknown new direction. I believe I can continue to produce high-quality, high-efficiency operators based on my understanding of engineering, upper-level model requirements, and underlying hardware. I also know I might love this new direction (or I might not), but the feeling of having my passion taken away is truly unpleasant. The quiet joy of sitting at my desk, calmly writing operators for an afternoon, may become a thing of the past this summer. I have to bury my talent in yesterday and become a mecha pilot. My hands have more gears, but my heart has lost some of its rhythm.To illustrate, imagine you’re a master knitter, especially skilled at creating various patterns and matching colors. Your sweaters are of excellent quality and beautifully patterned, attracting wealthy people from far and wide who commission you to knit for them, earning you a considerable sum. You thoroughly enjoy the feeling of sitting by the window, brewing a pot of tea, gazing at the green mountains, clear waters, cattle, sheep, and rising smoke outside, quietly knitting for an afternoon. But one day, someone invents a magical machine that, with just yarn and a pattern, can automatically knit sweaters of comparable quality and texture to your hand-knitted ones, and at a much faster speed. You know your competitors could easily reach your former level with this machine, so you feel compelled to use it too. You also know that, with the knitting skills you’ve accumulated over the past twenty years, even if everyone has a machine, your knitting speed and quality will still surpass your competitors. But the pleasure of listening to the rain by the window, threading a needle, and slowly passing the time was ultimately crushed by the roar of the machines.I know this is frustrating, but there’s no other way. I can keep my job, but I’ll most likely have to give up my old passions. I’m someone who keeps my rationality and emotion quite separate; I can be very rational when I need to handle problems, but sometimes I also show my emotional side. I remember crying my eyes out when I moved out of the rented apartment I’d lived in for a year, reluctant to part with the memories of the past. Saying goodbye to the era of handwritten arithmetic and human brain optimization today is undoubtedly even more cruel than this.I don’t know if any readers have similar feelings, but I think this is the only way things can be.What about people?While AI continues to advance, I also have some concerns:Are students today more likely to use AI to complete assignments, especially practical labs? Imagine having two choices: one is to spend eight hours painstakingly completing a lab, perhaps without even getting a perfect score; the other is to launch an AI model, costing only a few cents and a few minutes, and have the AI write perfect code. Which would most students choose?The above point leads to a severe deficiency in the engineering skills of many students, including the ability to organize code, build systems, anticipate future potential needs and address them in advance through design, and abstract thinking. Given the ever-increasing power of AI, are these “engineering skills” still necessary? Will these engineering skills gradually become obsolete like the old ability to “proficiently write x86 assembly language,” or will they remain valuable forever, like the ability to “understand the entire computer system from software to system to hardware”? If the latter, then it’s dangerous—a person with poor engineering skills, when paired with AI, can produce mountains of garbage several times more efficiently than before, thus sowing various seeds of disaster for the system and making the world even more chaotic.In the future society, will power be more important than technology or intelligence?These questions may require answers from the times themselves.ConclusionWith the development of AI, future societies may tend towards two extremes: communism and cyberpunk 2077. In the former, productivity is greatly liberated, and people’s living standards are significantly improved (I’ll stop here, otherwise I’m afraid it won’t pass censorship); while in the latter, a few tech companies control most resources, and only a very small number of people can use the most advanced AI and various technologies to achieve near-“mechanical ascension,” while most people can only use very weak AI. Social mobility will become increasingly difficult: you need the most powerful AI to cross social classes, creating a vicious cycle.Do you think that if Anthropic Corporation forever possesses the world’s most advanced AI, the future society will either become communist or like 2077? What do you think?Therefore, I still believe that cutting-edge intelligence should be available to everyone in an open and affordable way. I don’t trust Anthropic or OpenAI to do this, and I especially don’t want Anthropic to master the most advanced artificial intelligence or AGI; to exaggerate, the seriousness of that would be no less than Hitler acquiring atomic bomb technology before the Allies. This is why I chose to stay at DeepSeek: by researching powerful, fast, and inclusive artificial intelligence and open-sourcing it, we might be able to pull the world back from that 2077 timeline.May the world be at peace in the future. May all beauty be blessed.”
