而83岁的谢诺夫斯基，依然在实验室里追问那个问题。 也许没有人比他更适合回答今天AI缺失的那些碎片。他见证了神经网络从"异端"到"改变世界"的全过 程；他既懂物理学的简洁优雅，也懂生物学的复杂混沌；他和辛顿一起打开了AI的大门，又眼看着这 扇门后的世界变得越来越陌生。 1984年的一天，物理学家特伦斯·谢诺夫斯基和心理学家杰弗里·辛顿坐在实验室里，盯着黑板上的方程 发呆。那是AI的第二个寒冬，神经网络陷入僵局。人们都知道多层网络更强大，但没人知道怎么训练 它。 "如果我们把神经网络想象成一团气体呢？"谢诺夫斯基突然说。 这个疯狂的想法最终变成了玻尔兹曼机，这是一个用统计物理学重新定义"学习"的数学模型。它证明了 只要找到合适的能量函数，神经网络就能像气体从高温降到低温一样，自发地调整到最优状态。 这成为现代深度学习的理论基石之一。 但两人后续的志趣却互相有所偏离。辛顿发现了更实用的反向传播算法，带领深度学习走出寒冬，最终 迎来ChatGPT主导的AI时代。而谢诺夫斯基选择了回到神经科学实验室，用几十年时间解剖大脑的每一 个回路，试图回答那个最初的问题：大脑究竟是如何工作的？ 40年后，辛顿因玻尔兹曼机获得20 ...

2024年12月8日，诺贝尔物理学奖得主Hinton登台，发表了题为《玻尔兹曼机》的演讲。 当时，斯德哥尔摩大学Aula Magna礼堂内座无虚席，全球目光都集聚于此。 他深入浅出地分享了，自己与John Hopfield利用神经网络，推动机器学习基础性发现的历程。 如今，Hinton这个演讲的核心内容，于8月25日正式发表在美国物理学会（APS）期刊上。 论文地址：https://journals.aps.org/rmp/pdf/10.1103/RevModPhys.97.030502 1980年代，并存两种颇具前景的梯度计算技术—— 一种是，反向传播算法，如今成为了深度学习 核心引擎，几乎无处不在。 另一种是，玻尔兹曼机器学习算法，现已不再被使用，逐渐淡出人们的视野。 这一次，Hinton的演讲重点，就是「玻尔兹曼机」。 一开场，他幽默地表示，自己打算做一件「傻」事，决定在不使用公式的情况下，向所有人解释复杂的技术概念。 霍普菲尔德网络 找到能量最低点 什么是「霍普菲尔德网络」（Hopfield Network）？ Hinton从一个简单的二进制神经元网络入手，介绍了「霍普菲尔德网络」的核心思想。 每个神 ...

Group 1 - The concept of the Boltzmann Brain suggests that in an infinitely old and chaotic universe, random fluctuations could create a brain with complete memories and self-awareness without the need for a complex external world [1][2][3] - The probability of a Boltzmann Brain existing is argued to be higher than that of a low-entropy universe evolving into a complex structure, as the latter requires overcoming significant entropy increase [2][3] - This leads to the unsettling conclusion that human existence might be a fleeting phenomenon resulting from a random quantum fluctuation, challenging fundamental perceptions of reality [5][6] Group 2 - The discussion contrasts the Boltzmann Brain with Laplace's Demon, which represents determinism, suggesting that all thoughts and feelings are predetermined by physical laws [11][12] - Both perspectives imply that free will does not exist, whether through extreme randomness or absolute determinism [12][18] - Kant's philosophy attempts to reconcile these views by suggesting that true freedom exists beyond observable reality, yet this remains a scientific mystery [18][19] Group 3 - The insights from Boltzmann and Darwin regarding how order emerges from disorder provide a different perspective on evolution and consciousness [19][20] - Boltzmann's view redefines survival competition as a struggle for "negative entropy," indicating that life extracts order from its environment to maintain complexity [20] - This suggests that consciousness may be a product of evolutionary processes aimed at better perceiving the world and utilizing resources effectively [21][22] Group 4 - The exploration of consciousness requires a multidisciplinary approach, incorporating insights from cognitive science, philosophy, and neuroscience [40][42] - Various theories, such as Hofstadter's "strange loop," Turing's computationalism, and integrated information theory (IIT), challenge traditional notions of consciousness and its location [42][43][44] - These perspectives indicate that consciousness may not reside in a specific location but rather in the organization and flow of information within a system [46][47] Group 5 - The evolution of AI, particularly through models like the Boltzmann machine, reflects the potential for understanding consciousness through complex information processing [26][31][33] - The Boltzmann machine's design, which incorporates randomness and probabilistic learning, parallels the idea that consciousness may emerge from structured interactions within a chaotic environment [34][38] - This suggests that consciousness could be a result of cumulative processes rather than a singular miraculous event [38][39]