前OpenAI创始人称：大模型将从“堆芯片”转向“拼研究”

Core Viewpoint - The AI industry is approaching the limits of expanding computational power and needs to shift focus back to research for effective utilization of existing resources [2][5][6]. Group 1: Current Trends in AI - AI companies have previously focused on massive chip deployment and large-scale training data to expand computational power [3]. - The traditional belief that stronger computational power and more training data lead to higher intelligence in AI tools is being questioned [6]. Group 2: Insights from Industry Leaders - Ilya Sutskever, co-founder of OpenAI, emphasizes the need to find efficient ways to utilize existing computational power [4][7]. - Sutskever suggests that the industry must return to a research phase, supported by powerful computing, to advance AI development [5][6]. Group 3: Limitations of Current Approaches - The model of simply increasing computational power is nearing its limits, as data availability is finite and many institutions already possess substantial computational resources [6]. - Sutskever argues that merely scaling up computational resources will not lead to transformative changes in AI capabilities [6]. Group 4: Future Research Directions - There is a critical need for research focused on enhancing the generalization ability of models, allowing them to learn from minimal information, akin to human learning [7][8]. - The gap in generalization ability between AI models and humans is identified as a fundamental issue that requires attention [8].