Huang Renxun: In the next 10 years, AI computing power will increase by 1 million times

Source: Daily IC
    In two public videos, Huang talked in detail about how to deal with Nvidia‘s ups and downs, organizational structure management, AGI development, the growth potential of AI computing power, and future challenges.
   Huang said that Nvidia‘s success lies in the fact that all of Nvidia‘s work in the past 30 years has almost revolved around technology and the market, which is also the core of Nvidia: create technology before creating demand.
   When it comes to Nvidia‘s flat management, Huang mentioned that a cultural environment that does not believe that "information = power" is very important for employees to trust and empower. With just 30,000 employees, Nvidia is the smallest large company in the world, but each employee is given a lot of power.
    As the strongest seller in the era of generative AI, Nvidia‘s market value has exceeded $2 trillion. For AI trends, Huang Renxun said that AI will be the "defining technology" of this era, and in the future AI will become the "sovereignty" of a country, that is, a country must control its own digital intelligence and must be committed to controlling sovereign AI. One of the most recent breakthroughs in AI is deep learning, and another important breakthrough is the newly invented human feedback-based reinforcement learning (RLHF) technology for language models, which Nvidia has found a way to implement at the system level.
    Huang Renxun also believes that AI technology has narrowed the technological gap of mankind, the way of information processing in the future will be fundamentally changed from AI, generative AI will start from an information "seed", and the future of computing will be highly dependent on generation rather than retrieval. Moreover, AGI will pass human tests in the next five years, including professional tests such as the bar exam and gastroenterology, and AGI will come soon. However, he also acknowledged that achieving AGI from an engineer‘s perspective is difficult because engineers need clear goals, and AGI may still be some way off due to disagreements over how to describe how the human mind works.
   For the development of AI computing power, Huang Renxun believes that in the next 10 years, NVIDIA will increase the computing power of deep learning by 1 million times, so that AI computing equipment will continue to train, reason, learn, apply, and continue to improve, and the future will continue to transform super AI into reality.