AI专题“从上网到上算，由网络世界至虚拟现实”之十七：预计AI技术路径将由大模型切向具身智能

Investment Rating - The report maintains an investment rating of "Outperform the Market - A" [2][18]. Core Insights - The AI technology path is shifting from large models to embodied intelligence, with significant updates expected in 2024 and 2025 [1][11]. - The report outlines a framework of six computing platforms for AI applications, including humanoid robots, MR glasses, PCs, smartphones, vertical hardware, and the real world [1][11]. - The media sector is anticipated to rebound in Q4 2024, with publishing and gaming being favored by the market [1][14]. Summary by Sections 1. AI Technology Path Shift - Since the Spring Festival of 2023, AI has transitioned from ChatGPT to various models with regular updates, moving from a mapping market to one focused on performance and data-driven indicators [7][11]. - The large model's mapping market has ended, and the focus is now on the real-world application of AI technologies [7][11]. 2. Media Sector under AI Technology Iteration - The media industry is at a transitional phase, with AI expected to generate new content forms in 2025, utilizing AIGC for re-expressions in various media formats [13][14]. - The report identifies four key logic lines for the media sector, predicting a rebound in Q4 2024, with a focus on publishing and gaming sectors benefiting from AI-generated content [14]. 3. Investment Recommendations - The report lists several companies as key investment targets, including 分众传媒 (Focus Media), 中国电影 (China Film), and others across various segments of the media industry [15]. - The focus is on companies that can leverage AI for new content creation and those positioned for structural benefits in the evolving media landscape [14][15]. 4. Industry Dynamics - The report emphasizes the importance of embodied AI as the next stage following generative AI, with a complementary relationship between the two [11][12]. - The anticipated breakthroughs in embodied intelligence are expected to occur in 2025, with the current focus on overcoming challenges in sensing and protocol standardization [12][14].