Shulou(Shulou.com)11/24 Report--

CTOnews.com, Nov. 14, Sergei Edouov, engineering director of Meta generative artificial intelligence, recently shared his forecast of the reasoning needs of artificial intelligence at the Silicon Valley Digital Workers Forum. He believes that the reasoning needs of new artificial intelligence applications around the world next year can be met by the power generation of only two nuclear power plants if a language model of a reasonable scale is used.

Image source Pexels artificial intelligence reasoning refers to using the trained artificial intelligence model to carry out a variety of tasks in the actual scene, such as generating text, answering questions, identifying images and so on. Mr Edunov says he has estimated the amount of electricity needed for global reasoning next year through simple mathematical calculations. He assumes that 1 million to 2 million Nvidia H100 GPUs will be added worldwide next year, each with a power of about 1 kilowatt. If each processor runs 24 hours a day, 100000 "token" can be generated per person per day. He believes that this electricity consumption is reasonable on a human scale. Only two new nuclear power plants are needed around the world to provide enough electricity.

However, CTOnews.com noted that Edunov also pointed out that the development of artificial intelligence still faces some challenges and limitations, one of which is the amount of data. At present, the training of artificial intelligence model needs a lot of data, and the public Internet data is not enough to support the training of the next generation model. The next generation model may require 10 times the amount of data, which means more professional data, or more multimodal data, such as video, audio, and so on. Another challenge is the supply chain. Due to the shortage of global chip capacity, the speed of improvement of artificial intelligence models will also be affected. Therefore, researchers are trying to improve the efficiency of the model to reduce dependence on data and hardware. For example, Salesforce developed Blib-2, a technology that automatically resizes models and dynamically shrinks or expands models according to different tasks and resource requirements.

Industry experts generally believe that the language model will bring great value to enterprises within two years. Edouov predicts that within three or four years we will know whether current technology can achieve universal artificial intelligence.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.