DeepSeek Engineer's Growing Unease: More AI Work Means More AI Advancement, Fueling Catastrophic Fears

AI stands at the forefront of societal technological progress, with many now placing it above the internet and other groundbreaking discoveries. This prominence is understandable given its impact on our lives, particularly in the workplace. In a few short years, we've moved from generative AI that produced flawed images and videos to the sophisticated systems we see today. A DeepSeek engineer has entered a self-destructive cycle, stating that using his expertise to improve AI enables it to learn and automate, becoming even better and creating something that will escape our control.

An increasing number of people turn to artificial intelligence models for assistance with various tasks. If it were solely for complex or time-consuming activities, one might understand seeking AI as a last resort after attempting the task independently. However, people now directly consult ChatGPT and other models for easy answers or solutions to their problems, both personal and professional. As AI becomes increasingly precise and 'intelligent' in its responses, we witness growing dependence on it.

DeepSeek GPU Kernel Developer Acknowledges Self-Destructive Cycle from Training AI That Will Render Him Obsolete

The AI landscape offers a unique perspective when viewed through the eyes of those involved in its development. We now have the case of Liu Sengyu, a DeepSeek systems and ML engineer with expertise in GPU kernels. This developer possesses extensive knowledge of kernels and related fields. He finds himself in a 'self-destructive loop' because he believes he is using all his knowledge to train and improve AI, which in turn automates processes and accelerates everything.

From the developer's viewpoint, continuing his work means AI will advance more rapidly. He estimates that within 6-12 months, AI will be able to produce kernels at his level, making him replaceable. Naturally, he doesn't wish to lose his job, nor does he feel it is ethical to continue this path. The dilemma is that even if he stops contributing to AI development, many others will persist, and the situation will remain unchanged.

AI Developer Predicts a Future Where Hand-Coding Competitions Arise as Everything Becomes Automated

While the prospect of humans becoming easily replaceable can seem bleak, Liu still holds some hope. He believes that in his specific case, he won't be entirely replaced by artificial intelligence, but his work methods will shift. He will transition to directing and controlling AI agents. Personally, he feels this diminishes a significant part of what made his work engaging. Instead of individually programming and discovering what works and what doesn't, the work will largely be done, and he will only review minor errors.

Consequently, he is coming to terms with the idea that he will lose his passion for writing and optimizing kernels by hand. Indeed, AI agents have already begun learning low-level optimization techniques, leading the developer to believe that in the near future, manual programming without AI and machine assistance might evolve into a 'competitive sport'.