马斯克 vs 哈萨比斯 vs 杨立昆：谁定义的才是AI的真实未来？

Core Viewpoint - The debate surrounding AGI (Artificial General Intelligence) is highly polarized, with opinions ranging from optimistic predictions of imminent breakthroughs to skepticism about the feasibility of achieving AGI in the near future [2][21]. Radical Faction - The radical faction, represented by figures like Elon Musk and Dario Amodei, believes AGI is just around the corner, potentially within two years, with Musk predicting its arrival by the end of 2026 and collective intelligence surpassing human capabilities by 2030-2031 [3][4]. - Musk argues that AI capabilities double every seven months and that current models have a hundredfold potential for improvement, suggesting a chain reaction of advancements [3][4]. - Dario Amodei, while slightly more cautious, still maintains that AGI will emerge in the short term and emphasizes that safety measures should not hinder innovation [4][6]. Progressive Faction - The progressive faction, including leaders from DeepMind and OpenAI, advocates for a balanced approach to AGI development, emphasizing the importance of safety alongside innovation [10][11]. - Demis Hassabis defines AGI as a system with human-like cognitive abilities, stressing the need for understanding physical world interactions as a key to achieving AGI [12][13]. - This faction is more cautious about the timeline for AGI, with Hassabis estimating only a 50% chance of AGI being realized before 2030 [12][13]. Cooling Faction - The cooling faction, led by Yann LeCun, is skeptical about the current AI hype, arguing that existing technologies have fundamental limitations that make AGI unattainable in the short term [18][20]. - LeCun suggests that the term AGI has become a tool for hype and advocates for a focus on "advanced machine intelligence" instead, emphasizing the need for systems to understand physical world causality [18][19]. - This faction believes that the efficiency gains attributed to AI are overstated and that the current technology cannot replace human labor [20][21]. Conclusion - Despite differing views, all factions agree that pure text-based language models cannot achieve AGI and that developing a "world model" that understands physical world causality is essential [21]. - The year 2026 is seen as a pivotal moment for AI, marking a shift from scale-driven approaches to paradigm restructuring, with multi-modal models potentially becoming the core technology [21].