与OpenAI前科学家聊：GPT-5、创新与AI的欺骗性

Core Viewpoint - The article discusses the release of GPT-5 and the mixed reactions surrounding it, emphasizing the need for "no-goal innovation" as a pathway to true creativity and breakthroughs in AI [4][5][6]. Group 1: Innovation and AI Development - Kenneth Stanley and Joel Lehman advocate for the idea that significant innovations often arise unexpectedly, as evidenced by the development of ChatGPT [5][10]. - The concept of "no-goal innovation" suggests that pursuing novelty rather than specific objectives can lead to greater creative outcomes [11][12]. - The authors reflect on their experiences at OpenAI, noting that the company transitioned from a research-focused entity to a commercially driven organization, which may impact its innovative culture [24][25][28]. Group 2: Challenges in AI Research - The article highlights a perceived stagnation in AI research, with the authors expressing that the excitement of the field has diminished compared to earlier years [22][27]. - They discuss the risks of becoming overly focused on benchmarks and goals, which can lead to a deceptive sense of progress without genuine advancements in intelligence [44][45]. - The authors emphasize the importance of maintaining an open and exploratory approach to foster true innovation, especially in large organizations [36][39]. Group 3: Future of AI and Competition - The conversation touches on the competitive landscape of AI, particularly the emergence of new players and the evolving dynamics between companies like OpenAI and Google [33][34][35]. - The authors express concerns about the sustainability of "open exploration" in corporate environments, suggesting that financial pressures may hinder innovative efforts [38][39]. - They predict that the AI field will continue to evolve rapidly, with significant developments expected in coding and other applications [52][53].