乔治·霍茨

当前大模型的损失函数是最大化压缩，导致产出平庸，与追求AGI的目标背道而驰。

“如果你的损失函数仅仅是试图最大化压缩……它写的说唱听起来就像YouTube评论里的说唱。（If your loss function is just try to maximize compression... the raps that it wrote sounded like YouTube comment raps.）”

超级AI的终极风险不是直接消灭人类，而是打造令人无法抗拒的极致内容，操纵人类多巴胺系统使其沉迷、忘记生存需求。

“我们将打造一个你根本无法移开目光的TikTok。（We're going to build the TikTok that you actually can't look away from.）”

AI本身没有意图，真正的危险来自利用AI操控他人、集中权力的人类，而非机器。

“不是机器会那样做。是使用机器的其他人会那样对你。（It's not the machine that's going to do that. It's other humans using the machine that are going to do that to you.）”

霍兹对意识概念持怀疑态度，认为意识只是无神论者用来指代灵魂的词，甚至不认为自己有意识。

“我不认为自己有意识。意识……它就像是无神论者用来指代灵魂的词。（I don't think I'm conscious. Consciousness... it's just like a word that atheists use for souls.）”

众多AI芯片公司失败的根源在于软件栈糟糕：若无法在NVIDIA GPU上写出Torch级性能的软件栈，就更不可能为自研专用芯片做到。

“如果我无法在NVIDIA GPU上写出达到Torch性能水平的软件栈，那你也绝无可能为你的芯片写出这样的软件栈。（If I can't write a torch level performance stack on Nvidia GPUs, there's no way you're going to be able to write it on your chip.）”

Hotz 断言 MuZero 就是解决自动驾驶问题的方案，并将成为整个深度学习时代的典范论文。

“我认为 MuZero 将会被视为整个深度学习时代的典范论文。而 MuZero 就是解决自动驾驶问题的方案。（英文原文："I think MuZero is going to be looked back as the canonical paper of this whole deep learning era. And MuZero is the solution to self driving cars."）”

Hotz 长期极度看好加密货币的理念本身而非具体项目，将中本聪共识算法视为21世纪最伟大的创新之一。

“我长期极度看好加密货币。不是看好某个特定的项目，而是这种理念……中本聪共识算法是21世纪最伟大的创新之一。（英文原文："I'm extremely bullish on crypto long term. Not any specific crypto project, but this idea of... the Nakamoto consensus algorithm is one of the greatest innovations of the 21st century."）”

Hotz 认为特征工程类方法（如特斯拉数据引擎）终将被端到端学习方法取代并击败。

“我认为特征工程的方法最终总会被端到端的方法所取代和击败。（英文原文："I think feature engineering approaches will always be replaced and lose to end to end."）”

Hotz 澄清其 SXSW 演讲的重点不是字面上黑进模拟，而是将其视为理论物理问题：先构建大一统理论，再从中寻找漏洞。

“我在SXSW演讲的重点不是字面意义上的‘黑客攻击模拟’。我认为这是……理论物理的范畴。你想要一个大一统理论，但那就去构建一个大一统理论，然后寻找漏洞。（英文原文："The point of the South by talk was not literally to hack the simulation. I think that this is... theoretical physics. You want your grand unified theory, but then okay, build a grand unified theory, search for exploits”

Hotz 呼吁人类停止玩规则被操纵的虚假社会游戏，转而投入大自然写好规则的‘真实游戏’。

“我们应该谦卑一些……看看你把时间花在了什么上。人类的目标函数是什么？停止玩这些虚假的人类游戏，这里有一场真正的游戏。大自然已经写好了规则。（英文原文："We should humble ourselves... look at what you are spending time on. What is the goal function of humanity? Stop playing these fake human games, there's a real game here. Nature wrote the rules."）”