张亚勤院士：基础大模型最终不超过 10 个，十年后机器人比人多

Core Insights - The new wave of artificial intelligence (AI) is characterized by the integration of information intelligence, physical intelligence, and biological intelligence, which is fundamentally a fusion of atoms, molecules, and bits [4][6][31] - The transition from discriminative AI to generative AI, and subsequently to agent-based paradigms, marks a significant evolution in AI capabilities [6][31] - The future battlefield for AI over the next 5-10 years is expected to be the "Internet of Agents," where foundational large models will consolidate globally to no more than ten, replacing most current SaaS and apps [3][30][42] AI Development Trends - Generative AI is evolving towards agent-based AI, with agents showing a doubling in task length and achieving over 50% accuracy, aligning closely with human performance [39] - The scaling law in the pre-training phase has slowed, while post-training and inference stages are gaining momentum, with inference costs decreasing by tenfold over the past year [39][42] - The shift from information intelligence to physical and biological intelligence is underway, with significant milestones in autonomous driving and robotics anticipated in the coming years [40][55] Industry Structure and Opportunities - The foundational large models are likened to operating systems in the AI era, with a projected global landscape of no more than ten major players, split between China and the US [42][55] - Open-source models are expected to dominate the ecosystem, with an estimated 80% of platforms being open-source, fostering broader innovation and application [42] - The emergence of intelligent agents is not just a technological advancement but also signifies the formation of new economic structures, impacting enterprise management and product development [45][52]