对话启明创投：大模型三个月一迭代，没有永远的王者

Core Viewpoint - AI investment is transitioning from early exploration to a more strategic and systematic approach, reflecting a deep consideration of how AI can be effectively implemented [2]. Group 1: Large Models - The competitive landscape for large models is characterized by rapid technological evolution, with no single model maintaining a leading edge for more than three months [4]. - Companies are racing along a cost reduction curve, evolving from "usable" to "good" but have yet to reach "landmark applications" [4]. - AI combined with content generation is seen as the most likely area for the emergence of "super AI applications," potentially disrupting traditional internet connectivity models [4]. Group 2: Embodied Intelligence - Over 100 new companies focused on embodied intelligence have been established in the past two years, despite the saturation of the market [5]. - Entrepreneurs are driven by a deep-seated desire to connect robots with humans, viewing it as the ultimate technological form of connection [6]. - Significant challenges remain for the commercialization of robots, including engineering difficulties, data limitations, and the "uncanny valley" effect [6]. - The initial deployment of intelligent robots is expected in tasks such as picking, transporting, and assembling, which will help accumulate valuable operational data [6]. Group 3: AI Applications - The potential for "super AI applications" exists, but specific companies demonstrating this potential have yet to emerge [9]. - The cost of operating large models has decreased significantly, making the technical foundation for super applications more viable [9]. - The key to unlocking super applications lies in identifying which fields can create a strong demand for AI and which products can redefine user-platform boundaries [9]. Group 4: Investment Judgments - The distinction between products that are "shouted" into existence versus those that are "grown" is crucial for investment decisions [10]. - Reliable AI projects are characterized by tangible data and results, rather than just theoretical capabilities [10]. - There is a focus on smaller models in niche areas like AI for Science and AI safety, which may present significant opportunities in the next few years [10]. - A methodical approach to filtering out noise and identifying commercial viability is essential for successful investment in the AI sector [10].