丁宁：大模型是“智能基建”，资本与技术融合重塑AI版图

Core Insights - The event "Scientists Meet Investors" highlighted the significance of the fourth industrial revolution, emphasizing artificial intelligence (AI) and big data as core technologies for the future [1] - Professor Ding Ning from Xi'an Jiaotong University discussed the evolution of large language models (LLMs) and the shift from merely increasing parameter size to focusing on structural innovation and efficient training methods [2][3] Industry Trends - The industry is witnessing a transition from single-modal models to multi-modal integration, allowing AI to understand and generate information across various formats such as text, images, and speech [2] - The current trend in model training is moving towards continuous learning and parameter-efficient fine-tuning, enabling faster adaptation with lower computational costs [3] Capital and Technology Relationship - The relationship between capital and technology is crucial, where capital acts as a magnifier for technology, while technology drives capital efficiency [3] - High initial costs for training large models necessitate capital investment, but without technological insights, capital alone cannot drive industry upgrades [3] Global Comparison - The United States leads in top enterprises and computational resources, while China excels in research output, holding 41% of global AI papers and 69% of AI patents as of 2023 [3] - Despite advancements, computational power remains a bottleneck for AI development in China, with challenges such as model hallucinations and precision issues still needing resolution [3] Future Outlook - Future trends in AI development include multi-modal integration, parallel advancements in large-scale and lightweight models, embodied intelligence, and the exploration of artificial general intelligence (AGI) [4] - The concept of superintelligence, which refers to systems surpassing the smartest humans, remains a theoretical discussion and a potential future direction for AI [5]