At the Silicon Valley Digital Workers Forum, Sergey Edunov, director of generative artificial intelligence engineering at Meta, revealed a remarkable news: To meet the global growing demand for AI application reasoning next year, only two new nuclear power plants need to be added to power generation. This forecast has attracted widespread attention from the industry on AI energy consumption.
Edunov further explained that with the widespread use of Nvidia H100 graphics processors, the power consumption of global AI inference will be within a controllable range. He emphasized the importance of the verbal model generation work at a reasonable scale, believing that this will effectively control energy consumption while promoting the development of AI technology.
However, Edunov also pointed out that training larger language models will face insufficient data. He predicts that the next 3-4 years will be a critical period, when we will know whether current technology can implement general AI. This view triggered participants' deep thoughts on the future development of AI technology.
Edunov's speech not only reveals the current situation of the development of AI technology, but also provides important clues for future research directions. He called on the industry to focus on energy consumption while promoting the advancement of AI technology and ensure sustainable development.
In addition, Edunov also mentioned that Meta is actively exploring more efficient AI model training methods to reduce energy consumption. He believes that through technological innovation and cross-field cooperation, the future of AI technology will be brighter.
In general, Edunov's speech pointed out the direction for the development of AI technology, and also reminded us that while pursuing technological progress, it is necessary to consider its impact on the environment and energy. This view was widely recognized by the participants and also provided new ideas for future AI research.