Abstract:
Another masterpiece in the AI circle is born.
Recently, DeepSeek v4.1 was released, and its operator leader Liu Shengyu posted a long article titled "I had to bury my talent in yesterday." The content is too long, so the full text is at the end.
As soon as this article was published, it caused quite a stir at home and abroad. It has been translated into multiple languages on the Internet.
At the beginning of the article, he wrote: He wrote the main attention operator of v4.1, and the excellence of the model is an affirmation of his operator.
However, he said that he knew very well that in half a year or a year, the operators written by AI would most likely be as good as his, or even surpass him. Because AI can think about 300 tokens in one second, type out a line of commands in half a second, and write a code in 20 seconds, but he can't. AI can continuously improve in model depth, thinking intensity, tool call volume, and parallelism, but he cannot.
(Explanation: An operator is a specific calculation step in the AI model, and operator optimization is to write this step faster and more efficiently.)
This is not alarmist. He himself wrote: In just one year, AI has transformed from a little assistant that can only help him check documents, read code, and find bugs, to an "operator master" that can independently read CUDA, PTX and SASS codes, analyze the pause time of each instruction through professional tools, and then independently optimize operators.
Operator optimization is one of the most hard-core areas in AI infrastructure, and the threshold is extremely high. The AI can even do this kind of work, and the better it does it, the better.
But what makes people’s hearts sink the most are the next two sentences.
"Human beings have shown no hesitation in destroying themselves since ancient times."
"Of course I hope that I will not be revolutionized, but if I have to be revolutionized, I hope that the person who revolutionizes my own life is myself."
These two sentences are too harsh. It’s not that the AI wants to replace him, it’s the AI that he personally trained to replace him. Human beings have never hesitated to destroy themselves since ancient times, but this is not stupidity, but a kind of accelerationist helplessness. If you don't do it, others will do it. Rather than being revolutionized by others, it is better to revolutionize yourself.
As for his future, he wrote: "I will not be "unemployed," but I must "change jobs." I can still keep my job, but this may mean that I will never have the opportunity to do the job I once loved. I have to change from an operator engineer to an Agent mecha pilot."
"I had to bury my talent yesterday and become a mecha pilot. I have more gears in my hands, but less beats in my heart."
He also made an analogy. You are proficient in knitting sweaters, with unique patterns and colors. Rich people from all over the country come to ask you to knit sweaters. You also enjoy listening to the rain by the window, threading a needle and threading, and passing the time slowly. But one day, someone invented an automatic knitting machine. The quality was no worse than yours and the speed was much faster than yours. You have to use it. You know that with twenty years of craftsmanship, you are better than your peers even if there are machines. But the interest of handwork is eventually crushed by the roar of the machine.
At the end of the article, he talked about the future.
The development of AI may lead to two extremes: communism, which greatly liberates productivity; or Cyberpunk 2077, where a few technology companies control most of the resources and only a handful of people can use the most advanced AI. It becomes increasingly difficult to cross classes. You must first have the strongest AI to cross classes, forming an endless loop.
He also explained why he chose to stay at DeepSeek because he believes that cutting-edge intelligence should be available to everyone in an open and cheap way. He did not believe that some foreign AI companies could do this, and even used a very serious metaphor: allowing a certain closed-source company to monopolize the most advanced AI is no less serious than allowing extremist forces to master weapons before the Allies.
“DeepSeek researches powerful, fast, and inclusive artificial intelligence and open source it, which may bring the world back from the end of 2077.”
An engineer who writes an operator to replace himself, in the end, the last thing he thinks about is not his job, but whether the world will become cyberpunk. This kind of idealism may be the reason why DeepSeek can go all the way today.
He also left a message on the reward page at the end of the article:
Giving me rewards is not as cost-effective as charging DS tokens
.What do you think about this?
The original text is as follows:
Comments