DeepSeek V4.1 Operator Author on AI Impact: I Had to Bury My Talent in Yesterday

Beating AI News Flash: DeepSeek operator engineer Liu Sheng posted about the impact of AI on his work. He said the main Attention operator of DeepSeek V4.1 was written by him, but at the current pace of progress, in another six months to a year, AI's ability to write operators will most likely catch up with or even surpass his own.

A year ago, AI could only help him check documentation, read code, and find bugs. Now it can already read CUDA, PTX, and SASS, analyze the stall time of each instruction, and independently optimize operators. He expects that in the next step, AI will also design scheduling schemes on its own, evaluate performance, and complete the implementation.

He is very clear that the better he optimizes operators, the faster DeepSeek models will train and infer, and the faster AI will catch up with him. But even if he stops now, other companies' models will not stop. So he will still continue to make operators as good as possible, except that this competition may end up revolutionizing him out of his own job first.

He judges that he most likely will not lose his job, but may be forced to "change careers," from writing operators with his own hands to becoming a "mecha pilot" operating Agents. What he truly finds hard to accept is not the disappearance of his livelihood, but that he may never again have the chance to do the work he loves: "I have to bury my talent in yesterday."

The article also discusses at the end why he stays at DeepSeek. He advocates that the most cutting-edge AI should be provided openly and cheaply to everyone, and says bluntly that he does not believe Anthropic or OpenAI can do this. He worries that the strongest AI will ultimately be controlled by a small number of companies, further turning the technology gap into a gap in power and class.