Jensen Huang, CEO of Nvidia, recently discussed the ongoing challenges in artificial intelligence (AI) development, specifically addressing the issue of AI hallucinations. He stated that solving the problem of AI creating false or misleading information will take several more years. Huang emphasized that the focus should be on improving AI's trustworthiness, but also acknowledged the need for increased computational power to enhance AI's problem-solving abilities.
According to Huang, AI development is divided into three key stages: pre-training, post-training, and test time scaling. Pre-training involves AI ingesting massive amounts of data to form a foundational understanding. Post-training refines AI's abilities through reinforcement learning and feedback, while test time scaling, which Huang described as "thinking," allows AI to break down complex issues and simulate potential outcomes.
Despite advancements in these areas, Huang cautioned that the best AI responses today are still prone to errors, and further progress is needed to reach a point where the output is highly reliable. He stressed the importance of continuously increasing computational resources to achieve these advancements, as AI’s growing demands for processing power are projected to increase exponentially in the coming years.
Huang also reflected on Nvidia's contributions to the AI space, particularly in reducing the cost of computing by a factor of one million. This, he believes, has been a key factor in enabling the rapid growth of AI applications, as researchers now have access to powerful machines for machine learning at much lower costs.
Huang expressed that the industry’s focus must be on scaling up computational resources to meet the ever-growing demand for AI development, and that Nvidia’s role in this evolution will remain central to the future of artificial intelligence.
