Recently, former Google researcher Igor Mikhaylov stated at a tech summit: ‘The era of compute worship must come to an end.’ He argued that the AI industry has become overly reliant on scaling up computational resources to boost model performance, while neglecting algorithmic efficiency, energy consumption, and real-world applicability. This ‘bigger is better’ mindset not only wastes enormous amounts of energy but also reinforces technological monopolies, making it harder for smaller organizations to innovate.Mikhaylov, who contributed to early research on the Transformer architecture, urged a shift toward ‘green AI’ and ‘practical AI.’ He emphasized that genuine progress in artificial intelligence should stem from smarter architectural designs and more efficient data usage—not merely stacking GPUs. Models should deliver high cost-effectiveness for specific tasks rather than chasing ever-larger parameter counts.His remarks have sparked widespread debate. Supporters see this as a necessary course correction for AI development, while critics argue that scaling compute remains essential for exploring paths toward general intelligence. Regardless, as global attention turns increasingly toward sustainability, the AI field may be transitioning from compute-driven to intelligence-driven innovation.
近日，前谷歌研究员伊戈尔·米科洛夫（Igor Mikhaylov）在一次技术峰会上公开表示：‘算力崇拜的时代该结束了。’他指出，当前人工智能行业过度依赖提升计算资源来推动模型性能，却忽视了算法效率、能源消耗和实际应用场景的适配性。这种‘越大越好’的思维不仅造成巨大的能源浪费，也加剧了技术垄断，使中小机构难以参与创新。米科洛夫曾是Transformer架构早期研究的重要参与者之一，他对当前大模型竞赛提出反思：真正的人工智能进步应来自更聪明的架构设计、更高效的数据利用方式，而非单纯堆砌GPU。他呼吁业界转向‘绿色AI’与‘实用AI’，强调模型应在特定任务中实现高性价比，而非一味追求参数规模。这一观点引发了广泛讨论。支持者认为这是对当前AI发展路径的必要纠偏；批评者则指出，在通用智能尚未实现前，扩大算力仍是探索有效路径的重要手段。无论如何，随着全球对可持续发展的关注加深，AI行业或将迎来从‘算力驱动’向‘智慧驱动’的转型。
原创文章，作者：admin，如若转载，请注明出处：https://avine.cn/12184.html