DeepSeek has released its v4.1 model, featuring core attention operators developed by Liu Sheng. In a widely circulated article, the lead engineer acknowledged that AI is rapidly reaching a level of proficiency in operator optimization that matches or exceeds human capability. He noted that AI can now independently analyze CUDA, PTX, and SASS code, as well as optimize instructions with high efficiency.
Liu Sheng stated that while his job is not immediately at risk, he must transition from a traditional operator engineer to an 'Agent pilot.' He emphasized that the manual craftsmanship of coding is being replaced by machine automation. This shift reflects a broader trend where engineers are effectively training the very systems that automate their specialized tasks.
The engineer expressed concerns regarding the future of AI development, contrasting the potential for massive productivity gains with the risk of a 'Cyberpunk 2077' scenario, where a few companies control advanced technology. He remains at DeepSeek because he supports the company's commitment to open-source, affordable, and accessible artificial intelligence.
Liu Sheng argued that preventing the monopolization of advanced AI by closed-source entities is critical. He believes that DeepSeek's open-source approach serves as a counterweight to potential technological dominance by foreign firms. At the end of his article, he encouraged supporters to invest in DeepSeek tokens rather than donating to him personally.
