对话原旷视联创唐文斌：现在很多具身订单都是“伪商业化”

Core Viewpoint - The article discusses the emergence of a new robotics company, Dexmal, founded by former executives of Megvii, aiming to explore the potential of embodied intelligence in robotics, which is seen as a continuation of their previous ambitions in AI technology [4][5][14]. Group 1: Company Background and Vision - Dexmal was established in March 2025, with the goal of revisiting and advancing the robotics ambitions that were not fully realized during the previous "AI four little dragons" era [5][18]. - The founder, Tang Wenbin, emphasizes the importance of creating functional value in robots, moving beyond mere emotional or display capabilities [8][9]. - The company adopts a unique approach by focusing on developing models first, which will then inform the hardware design, contrasting with other companies that prioritize hardware development [9][10]. Group 2: Industry Insights and Trends - The robotics industry is still in its early stages, lacking a consensus on technology routes, product forms, and commercial scenarios, and has not yet experienced a "ChatGPT moment" [11][12]. - Tang believes that the industry will eventually consolidate, but it is likely to retain a larger number of players compared to the large model sector due to the complexity of robotics, which involves various engineering disciplines [36][37]. - The article identifies three main application scenarios for robots: entertainment and research, industrial and logistics, and service-oriented interactions with consumers [41][42]. Group 3: Challenges and Opportunities - A significant challenge in transitioning from emotional value robots to functional value robots lies in the complexity of the robots' "intelligence" or cognitive capabilities [54]. - The current market is characterized by a prevalence of "pseudo-commercialization," where many robots are sold for demonstration purposes rather than for practical, ongoing use [48][51]. - The article highlights that many existing robots are not yet capable of sustained operation in real-world environments, indicating a gap between demonstration and practical application [53].