AGI根本不存在？图灵奖得主怒斥“这是谎言”，诺奖得主回怼“完全错误”！｜万亿美金的路线分歧：杨立昆和哈萨比斯，谁手里才是真正的藏宝图？

wow

https://www.youtube.com/watch?v=wu8dX9dFYl0

Dec 25, 2025 #AGI #YannLeCun #DemisHassabis

通用智能（AGI）竟然是一个谎言？图灵奖得主杨立昆（Yann LeCun）怒斥“根本没有AGI这种东西”，而诺奖得主、AlphaGo之父哈萨比斯（Demis Hassabis）则回怼“完全错误”！这场发生在AI金字塔尖的激烈交锋，不仅仅是定义之争，更是一个决定了数万亿资金流向和人类未来命运的巨大道岔。本期视频，我将带你深度拆解这场关于“特化”与“通用”的巅峰辩论，看懂两位顶尖科学家如何用黑猩猩、F1赛车和图灵机理论，重新定义“智能”的本质。 Is Artificial General Intelligence (AGI) a lie? Turing Award winner Yann LeCun claims "There is no such thing as AGI," while Nobel Laureate and AlphaGo creator Demis Hassabis fires back, calling him "Plain incorrect." This clash at the very summit of AI isn't just about semantics; it's a critical fork in the road determining the direction of trillions in investment and the future of humanity. In this video, we deconstruct this peak debate on "Specialization vs. Generalization," exploring how chimps, F1 cars, and Turing Machines are being used to redefine the essence of "intelligence." ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 📄 核心内容 & 关键词 | Key Content & Keywords: 特化 vs 通用 (Specialization vs. Generalization): 我们深入分析了杨立昆的核心观点——智能本质是为生存而“特化”的高效工具，以及哈萨比斯的反驳——人类大脑具备跨领域的“适应性”和“灵活性”。 We analyze LeCun's core argument that intelligence is a highly "specialized" tool for survival, contrasted with Hassabis's counter-argument regarding the human brain's cross-domain "adaptability" and "flexibility." 熵滤镜 (Entropy Filter): 为什么我们觉得世界是可以理解的？视频揭示了大脑如何像“滤镜”一样过滤掉宇宙中绝大多数无序信息（熵），只保留利于生存的结构。 Why does the world seem understandable? We reveal how the brain acts as a "filter," ignoring the vast majority of the universe's entropy and retaining only the structures necessary for survival. 近似图灵机 (Approximate Turing Machines): 哈萨比斯如何利用计算机科学的基石概念，证明大脑和现代AI在底层架构上具有“通用性”，而目前的局限只是资源问题。 How Hassabis uses the foundational concept of Turing Machines to argue that the brain and modern AI are fundamentally "general" in architecture, and current limitations are merely resource constraints. 两条路线的决裂 (The Roadmap Split): 这场辩论决定了AI研发的未来：是构建由无数“专家AI”组成的协作生态系统（杨立昆派），还是坚持“规模即一切”，打造全能的巨型模型（哈萨比斯派）？ This debate dictates the future of AI development: building an ecosystem of collaborative "expert agents" (LeCun's camp) or sticking to "Scale is All" to create a singular, omnipotent model (Hassabis's camp)? ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬