DeepSeek founder provides clue on start-up’s AI priorities in new technical study
DeepSeek is prioritizing the development of more efficient AI models, as indicated by a new technical study focusing on "native sparse attention" (NSA). This system, detailed in a study co-authored by founder and CEO Liang Wenfeng, aims to streamline the processing of vast amounts of data by AI, reducing costs without compromising performance.
The study, published just before Liang attended a symposium with President Xi Jinping, highlights DeepSeek's commitment to research and innovation. The NSA system optimizes computing hardware to accelerate inference – the process by which an AI model recognizes patterns and makes predictions on unseen data. This focus on efficiency follows DeepSeek's previous breakthroughs in developing advanced open-source AI models like V3 and R1 at significantly lower costs than industry standards.
via SCMP Full Text Feed
DeepSeek is prioritizing the development of more efficient AI models, as indicated by a new technical study focusing on "native sparse attention" (NSA). This system, detailed in a study co-authored by founder and CEO Liang Wenfeng, aims to streamline the processing of vast amounts of data by AI, reducing costs without compromising performance.
The study, published just before Liang attended a symposium with President Xi Jinping, highlights DeepSeek's commitment to research and innovation. The NSA system optimizes computing hardware to accelerate inference – the process by which an AI model recognizes patterns and makes predictions on unseen data. This focus on efficiency follows DeepSeek's previous breakthroughs in developing advanced open-source AI models like V3 and R1 at significantly lower costs than industry standards.
via SCMP Full Text Feed
