“按需思考”的GPT-5引发争议，但这可能是AI的未来

Core Viewpoint - OpenAI's release of GPT-5 has turned into a public relations and trust crisis due to user dissatisfaction with the model's performance and the introduction of a new routing technology that users feel has stripped them of control [1][2]. Summary by Sections Release and Initial Reactions - The launch of GPT-5 was expected to solidify OpenAI's leadership in the AI field, but it faced backlash from users who mourned the loss of their favorite model, which served multiple roles [1]. - Critics, including Gary Marcus, labeled GPT-5 as "overhyped" and "lackluster," indicating a decline in user satisfaction [1]. Technical Aspects of GPT-5 - GPT-5 utilizes a routing technology that automatically allocates tasks to different sub-models, which has led to performance inconsistencies [1][3]. - Users were surprised to learn that GPT-5 is not a single model but a network of multiple models, some of which are less capable and cheaper [1][3]. User Response and Company Actions - In response to the backlash, OpenAI re-enabled the earlier model GPT-4o for professional users and promised to fix routing issues and improve system stability [2]. - Anand Choudhury from FirstQuadrant commented on the dual nature of routing technology, highlighting its potential for both magic and failure [2]. Future of Model Routing Technology - Experts believe that model routing technology will become standard due to the limitations of single models and the economic benefits of reusing older models [3][4]. - The physical limitations of GPU memory and the challenges in scaling models further support the need for routing technology [4]. Historical Context and Criticism - The concept of model integration is not new, having emerged around 2018, but the current implementation in GPT-5 has been criticized for being overhyped [5]. - William Falken noted that the improvements from GPT-4 to GPT-5 are minimal compared to previous iterations, which has contributed to user dissatisfaction [6]. AGI Aspirations and Industry Perspectives - The debate surrounding model routing has led to skepticism about the imminent realization of Artificial General Intelligence (AGI), with some experts questioning the feasibility of achieving AGI through current models [6][7]. - The industry recognizes that while routing technology offers advantages, the path to AGI remains complex and uncertain, with a need for a balance between specialized models and unified large models [7].