Anthropic's Fable Backlash, Nationalizing AI & Broken Elections

"All right, everybody. Welcome back to the number one podcast in the world... We're doubling down with the original quartet. Everybody's here for a big week, lots of debates, and we gotta start with: Anthropic released a Macy's-level model with some interesting guardrails, and Dario is back to blogging."

"好了各位,欢迎回到全世界第一的播客……我们这一期是原班四人组全员到齐,这周料很足、辩论很多。开场就得聊它:Anthropic 发了一个堪称'重磅级'的模型,带着一堆有意思的护栏,而且 Dario 又开始写博客了。"

"The model's called Fable 5. No idea why. Released on Tuesday and it tops nearly every benchmark. The tokens, however, cost twice as much as Opus 4.8, but it should use fewer tokens overall because it's theoretically that much better."

"模型叫 Fable 5,不知道为什么取这名。周二发布,几乎屠了所有 benchmark。不过它的 token 价格是 Opus 4.8 的两倍——但理论上因为它聪明那么多,总的 token 用量会更少。"

"Here's what got the big developer backlash. While using Fable, Anthropic stores all the prompt data you enter for at least 30 days — a bit of a privacy issue. And if Fable 5 detects you're doing frontier AI research — in other words, using their model to make a better, competitive model — they were downgrading you, but not telling you. And this was buried deep in a 319-page document."

"真正点燃开发者怒火的是这个:用 Fable 的时候,Anthropic 会把你输入的所有 prompt 数据至少存 30 天——这有点隐私问题。而且如果 Fable 5 检测到你在做前沿 AI 研究——说白了就是用它的模型去造一个更好的、有竞争力的模型——它就会偷偷给你降级,还不告诉你。这条规定埋在一份 319 页的文档深处。"

"X freaked out over this, but they've since walked it back a bit — in Wired they said 'we're changing Fable 5 safeguards for frontier LLM development to make them more visible.' Chamath, what's your take? Are they being thoughtful, or are they being drama queens?"

"X 上为这事炸了锅,后来他们稍微往回收了一点——在《Wired》上表态说'我们正在调整 Fable 5 针对前沿 LLM 研发的安全措施,让它更透明'。Chamath,你怎么看?他们是想得周到,还是在演苦情戏?"

"It's a really incredible model, and kudos to these guys for continuing to push the boundaries of the closed frontier lab models. They're firing on all cylinders. But I think it creates a pretty obvious risk, and that risk is twofold. One: Anthropic has essentially shown their hand — they will increasingly take in prompts, evaluate them, and decide what to do with them before they generate output to you."

"这模型确实牛,得给他们点个赞,持续在闭源前沿实验室这条线上推边界,火力全开。但我觉得它带来一个相当明显的风险,而且是双重的。第一:Anthropic 基本上把底牌亮出来了——他们会越来越多地把你的 prompt 收进去、评估一遍,然后在给你输出之前先决定该拿它怎么办。"

"If you're a person, you should now think there's a risk of censorship. If you're a company, it's almost a nonstarter — because you could trip one of these things without even knowing it. A downstream scientist using the cloud APIs could trip it, a business executive could trip it, a person doing molecular research could trip it. And all of a sudden you'll get cut off from a very important source of business differentiation."

"如果你是个人,你现在就该意识到有被审查的风险。如果你是公司,这几乎是直接出局——因为你可能在毫不知情的情况下就触发了它的某条红线。下游一个用 cloud API 的科学家可能触发它,公司里一个高管可能触发它,一个做分子研究的人可能触发它。然后突然之间,你赖以建立差异化竞争力的一个重要来源就被切断了。"

"So companies need to start underwriting this next phase of AI: how do I have control? Who am I allowing to learn off all this information? Do I want single-point-of-failure risk with AI? The answer is you need broad diversity and a better-managed governance approach. One thing I'll say about Anthropic — they tell the truth."

"所以公司得开始认真评估 AI 的下一个阶段:我怎么保有控制权?我允许谁从这些信息里学习?我愿不愿意在 AI 上承担单点故障的风险?答案是,你需要足够广的多样性,和一套管理得更好的治理方式。关于 Anthropic 我得说一句公道话——他们说的是真话。"

"It's just that the truth sucks when you actually take it and you eat it and it's in your tummy and you're like, no, this is not good. So there's the censorship risk on one hand, and the governance business risk for enterprises on the other. Both are not good."

"只不过这个真话,等你真把它吞下去、它在你胃里翻腾的时候,你会觉得'不行,这玩意儿不对劲'。所以一边是审查风险,另一边是企业的治理与经营风险。两个都不是好事。"

"Freeberg, you're running your own company now. Do you worry about getting rug-pulled — investing your time into one of these platforms, then having them constrain your use, or take what you're doing at Ohalo, put it into their model, and make it available to other people?"

"Freeberg,你现在自己在经营公司。你会不会担心被'抽地毯'——你把时间押注在某个平台上,结果它转头限制你的使用,或者把你在 Ohalo 做的东西吸进它的模型、再开放给别人用?"

"There are things I'm concerned about, but we've decided to throw caution to the wind, because we do a lot of proprietary work in genomics. We look at gene variants, estimate the impact a variant might have on a living organism — we work in plants, in agriculture, we do RNA guide design for gene editing. These models have been incredibly valuable. Six months ago we could very cleanly design a genetic construct to make a specific protein."

"有些事我确实担心,但我们决定豁出去了,因为我们在基因组学上做了大量专有研究。我们研究基因变体,估算某个变体对活体生物的影响——我们做植物、做农业,做基因编辑用的 RNA 向导设计。这些模型一直极其有用。半年前我们还能干干净净地设计出一段基因构件,用来造出某种特定蛋白质。"

"Over the last couple of weeks, they've begun to restrict the ability to do that work — the premise being some sort of bio-weapon risk people have theorized. As a result, we're losing the capacity to use these models for this important scientific work. So we're likely going to end up needing to use open source models and run them locally ourselves."

"最近这几周,他们开始限制做这类工作的能力——理由是有人推演出某种生物武器风险。结果就是,我们正在失去用这些模型做这项重要科研的能力。所以我们大概率最后只能转向开源模型,自己在本地跑。"

"And what are the best open source models today? They're Chinese. The American open source models are not as good as the Chinese open source models. So the restrictions Anthropic and others are putting on themselves and the industry are forcing a lot of companies to go get open source Chinese models and run them. We're seeing it across the landscape, with startups and large-scale enterprises. Everyone's making that move."

"那今天最好的开源模型是谁的?是中国的。美国的开源模型不如中国的开源模型。所以 Anthropic 这些公司给自己、给整个行业上的这些限制,正在逼着一大批公司去搞中国的开源模型、自己跑起来。从初创到大型企业,整个生态里我们都在看到这个趋势。所有人都在往这个方向走。"

"I predict your next card to turn over is you'll start making your own models — you'll take all this data you have..."

"我预测你接下来要翻的下一张牌,就是你会开始造自己的模型——拿你手里这一堆数据……"

"Exactly right. We'll start with the core model, combine it with our data, and then we'll have our own genome language model, our own prediction model that we use internally. A lot of folks are feeling pressure from Anthropic and from politicians repeating the scary words Dario and others are saying. They're trying to force either natural or politicized enforcement on the model providers — in a way that ultimately benefits Chinese open source providers. That's scary, because we're going to damage our own economic viability."

"完全正确。我们会从一个基础模型起步,把它和我们的数据结合,然后就有了自己的基因组语言模型、自己的预测模型,内部自用。很多人正同时承受两股压力:一股来自 Anthropic,一股来自那些把 Dario 们的吓人言论挂在嘴边的政客。他们正试图对模型提供方强加某种自然的或政治化的管制——而这种做法最终只会便宜了中国的开源模型提供方。这很可怕,因为我们这是在自毁经济竞争力。"

"You can't just stop AI. By trying to stop it through political action, you're fundamentally going to give someone else the advantage, because the AI isn't going to go away. You'll just force someone else's hand to now have an unfair advantage, because you're unfairly restricting your own models and your own access to them."

"你没法直接把 AI 叫停。你想靠政治手段去阻止它,本质上只是把优势拱手让给别人,因为 AI 不会就此消失。你只会逼出别人,让对方反而拿到不公平的优势——只因为你在不公平地限制自家的模型、限制自己对这些模型的使用。"

"Sacks, what's your take on the chessboard? We've got the business case and the local case, but there are a lot of issues with Anthropic essentially throwing up every red flag to get regulated — which is our second story, Bernie Sanders seizing the equity of these companies. They're saying 'look over here, we're starting a fire, regulate the hell out of us.'"

"Sacks,从棋盘全局看你怎么想?我们已经听了商业视角和本地部署视角,但 Anthropic 这边问题不少——它几乎是举着每一面红旗求监管(这正好是我们第二个话题:Bernie Sanders 要没收这些公司的股权)。他们就像在喊'快看这边,我们点了把火,你们赶紧使劲管我们'。"

"Eight months ago I said Anthropic was engaged in a very sophisticated regulatory capture campaign based on fear-mongering. People thought that was a very IC take. But eight months later, you're hearing a lot of people say it — it's almost becoming a new consensus."

"八个月前我就说过,Anthropic 在搞一场非常精密的、靠贩卖恐惧来推动的监管捕获。当时大家觉得这是个很'非主流'(IC)的看法。但八个月后,你会听到一大堆人在说同样的话——它几乎正在变成新的共识。"

"What your summary didn't capture, JCal, is the sense of violation of trust and the outrage in the developer community. It's not just mandatory surveillance — remember, this is the company that said it was against government surveillance. They're now retaining for 30 days every prompt and every output you send to one of these Mythos-class models. No exceptions. Even enterprise customers who signed zero-data-retention agreements don't have a choice."

"JCal,你的总结里漏掉了一点——那种信任被践踏的感觉,以及开发者社区里的愤怒。这不只是强制监控的问题——别忘了,这可是一家曾经公开反对政府监控的公司。它现在要把你发给这些 Mythos 级模型的每一个 prompt、每一次输出都保留 30 天。没有例外。连签了'零数据留存'协议的企业客户都没得选。"

"And it's not just the prompts and output — it's all the context you share. All these agent platforms store your memories, your files, your data, and pass them to the model in giant context windows. Anthropic says it'll keep all of that, to build a profile on you, classify you, and determine what capabilities it unlocks for you."

"而且不只是 prompt 和输出——是你分享的全部上下文。这些 agent 平台把你的记忆、文件、数据全存下来,塞进巨大的 context window 里喂给模型。Anthropic 说这些它统统保留,用来给你建画像、给你归类,再据此决定给你解锁哪些能力。"

"The thing that created the most outrage is that they would degrade the product — nerf the model — if Anthropic in its sole discretion decided you're not worthy of that level of information. They're creating a new level of AI haves and have-nots. The narrow piece they walked back: for machine learning, AI research, chip design, they'd kick you to a lesser model and not tell you."

"最让人火大的是:如果 Anthropic 单方面认定你'不配'拿到那个级别的信息,它就会把产品做降级处理——给模型阉割(nerf)。他们这是在制造一个新的等级:AI 的有产者和无产者。他们松口收回的那一小块是:在机器学习、AI 研究、芯片设计这些领域,他们会把你踢到一个更差的模型,而且不告诉你。"

"That feels anti-competitive, right?"

"这感觉就是反竞争啊,对吧?"

"Completely anti-competitive. They'd even rewrite your prompt in the background, give you a nerfed answer, not tell you what they were doing, still charge you for the product you thought you were getting, and never tell you that you were not getting frontier-model capability. They were actually misleading their users."

"百分百反竞争。他们甚至会在后台偷偷改写你的 prompt,给你一个被阉割过的答案,不告诉你他们做了什么,照样按你以为买到的那个产品收你钱,而且永远不告诉你其实你拿到的根本不是前沿模型的能力。这就是在赤裸裸地误导用户。"

"The narrow piece they walked back: they'll now disclose when they downgrade you. But they're still downgrading people. Someone showed that simply asking about mitochondria got them downgraded. When Ben Thompson from Stratechery asked a straightforward question about the relationship between cancer risk and GLP-1s, he got kicked out. These guys have a very expansive view of who should be downgraded."

"他们往回收的那一小块是:以后给你降级时会告知你。但他们还是在给人降级。有人晒出来,单纯问一句线粒体的问题就被降级了。Stratechery 的 Ben Thompson 问了个很直白的问题——癌症风险和 GLP-1 类药物的关系——结果就被踢出去了。这帮人对'谁该被降级'的判定范围大得离谱。"

"This is a vision of where it's all headed: powerful big-tech companies deciding whether you're an AI have or have-not, then censoring the output you receive based on the profile they build on you. That is very Orwellian."

"这就是一切走向的缩影:强大的科技巨头来裁定你是 AI 的有产者还是无产者,然后根据它给你建的画像,审查你能收到的输出。这非常奥威尔。"

"There's something more subtle but equally insidious: they'll start to pick which corporations they want to benefit. If they have a strategic deal with Novartis and not Lilly, and Novartis has a competing GLP-1 drug, now there's an impetus to shape how people get information. If they have a deal with JP Morgan but not Citibank, you'll just never know — you'll ask about mortgage rates and get all kinds of stuff you don't understand."

"还有一层更微妙、但同样阴险的风险:他们会开始挑选自己想让谁受益。假设他们跟 Novartis 有战略合作、跟 Lilly 没有,而 Novartis 又有一款跟 Lilly 竞争的 GLP-1 药——现在就有动机去左右人们获取信息的方式了。如果他们跟摩根大通有合作、跟花旗没有,你根本无从察觉——你问个房贷利率,得到的全是你看不懂的东西。"

"And it's not clear they'll capture the actual fingerprint of what they did. If you believe something was done in a shady way, you should be able to go to them — or a regulator should — and say 'show me the trace of that exact model run, convince me it wasn't nerfed on purpose.' There are all kinds of ways to hide the cheese on that. That's what's scary."

"而且我不确定他们会不会真的把自己做过什么的'指纹'记录下来。如果你怀疑某件事是被暗箱操作的,你应该能去找他们——或者监管者去找他们——说'把那一次模型运行的 trace 调出来给我看,说服我它不是被故意阉割或操纵的'。但要把这块奶酪藏起来,办法多得是。这才是吓人的地方。"

"Let me give my free-market view: should that even matter, if there are a bunch of models out there and we don't restrict them? Let the markets compete. We had paid inclusion in the early days of search engines — companies paid to be in the results. Those engines sucked for consumers, so everyone said 'this isn't as good as Google' and went to the better provider. Isn't that what's gonna happen here, if we keep the government out of it?"

"我说说我的自由市场视角:如果外面有一堆模型、我们又不去限制它们,这事还要紧吗?让市场去竞争就好了。搜索引擎早期有过'付费收录'——公司花钱才能进结果页。那些搞这套交易的引擎对用户体验很差,于是大家说'这玩意儿不如 Google',就都跑去更好的那家了。只要我们别让政府插手,这里不也会是同样的结局吗?"

"You're missing one piece. At the same time Anthropic was doing mandatory surveillance and nerfing its models, Dario wrote a new blog post saying transparency was no longer good enough — that we need a new regulatory agency, like an FAA or FDA, to approve all models. Your premise (you could go elsewhere) is correct, yes. But at the same time they're engaging in this anti-competitive behavior, they want to restrict your options. It's not enough for them to win in the free market — they're calling on the government to stop potential competitors."

"你漏了一块。就在 Anthropic 搞强制监控、给模型做阉割的同时,Dario 写了篇新博客,说'透明已经不够了'——我们需要一个新的监管机构,像 FAA(联邦航空局)或 FDA(食药监局)那样,去审批所有模型。你的前提(你可以换一家)本身没错。但问题在于,他们一边做着反竞争的事,一边又想堵死你的备选项。光在自由市场赢还不够——他们在呼吁政府去拦住潜在的竞争对手。"

"Why would they want to get regulated? Simple — they don't. They want to apply that regulation to open source, which is impossible to regulate. So it's a way to scuttle and sandbag open source models — a preemptive strike against token-maxing on your local machine."

"他们为什么会想被监管?很简单——他们不想。他们想把这套监管套到开源头上,而开源根本没法被监管。所以这是个绝佳的办法去搞垮、去拖死开源模型——一记针对'你在本地机器上自己拉满算力'的先发制人。"

"Two years ago I bought 2,000 acres in Arizona with a partner. We got it zoned and approved to build a 2-gigawatt data center. I thought this was the moment you flip it to the Blackstones, Brookfields, Googles of the world and let them develop it. But I've come to the conclusion that I can't — because unless we direct large swaths of compute exactly the way Freeberg, Sacks and Jason are describing, we won't have open source."

"两年前我和一个合伙人在亚利桑那买了 2000 英亩地。我们办下了分区许可和审批,可以建一个 2 吉瓦的数据中心。我本以为这正是那种时刻:转手卖给黑石、博枫、Google 这些巨头,让他们去开发。但我现在的结论是——我不能这么干。因为除非我们能像 Freeberg、Sacks 和 Jason 说的那样,把大片算力精准地导向开源,否则我们根本不会有开源。"

"When you look at the long tail of access to open source models, it's relatively minuscule. The overwhelming majority of megawatts still goes to the big models. So I'm concluding we may just get dragged into having to build the 2 gigawatts, and I just put in an offer for another gig elsewhere — because if you could take 3 gigawatts and create deep, liquid access to open source, we as a community are gonna have to do it."

"你去看开源模型那条接入的长尾,占比其实小得可怜。绝大多数的兆瓦还是流向了那些大模型。所以我得出的结论是:我们可能会被硬拽进去,不得不把那 2 吉瓦建起来——我还刚刚在另一个地方出价又拿了 1 吉瓦。因为如果你能凑出 3 吉瓦,真正给开源造出又深又有流动性的算力入口,那我们作为一个社区就必须得做。"

"Now, I don't have the money — a gigawatt now costs $100 billion. When I started this project it was 4 or 5 billion; it's increased 20x. So to develop all 3 gigawatts I'd have to come up with $300 billion. I don't have $300 billion. So what are we all supposed to do? And if you get the rug-pull on the regulatory side too, we'll be stuck with one class of models and one set of rules."

"问题是,我没那个钱——现在一吉瓦要 1000 亿美元。我刚开始做这个项目的时候才 40 到 50 亿,如今涨了 20 倍。所以要把这 3 吉瓦全建出来,我得掏 3000 亿美元。我显然没有 3000 亿。那我们到底该怎么办?如果监管这边再给你来一记抽地毯,我们就会被困死在唯一一类模型、唯一一套规则里。"

"We're already seeing China race ahead in biotech, in material science, in new industrial systems. The capabilities that are going to be lopsided will create an unfair advantage — and that's when our workforce and our economy are truly at risk."

"我们已经看到中国在生物科技、材料科学、新型工业系统上一路狂奔。这种能力上的失衡会制造出不公平的优势——到那时,我们的劳动力和经济才是真正岌岌可危。"

"Hey Freeberg, isn't it true you guys use an open source model for genomics? Didn't the Collisons put out one — Evo 2?"

"Freeberg,你们做基因组学其实是在用一个开源模型吧?Collison 兄弟不是出了一个嘛——Evo 2?"

"It's great. The Arc Institute ingested all the world's genomic data they could get and built a genome language model. In our plant-breeding program, we're trying to figure out whether a particular gene variant is good or bad. You feed it in, and the model looks at the order of the DNA letters and tells you 'that's a good set of letters or a bad set' — almost like 'is that good English or bad English?' It's figured out the language of DNA, because it sees the probability of those letters being in a row across other organisms."

"非常好。Arc Institute 把他们能拿到的全世界基因组数据都喂了进去,造出一个基因组语言模型。在我们的植物育种项目里,我们要判断某个基因变体到底是好是坏。你把它输进去,模型会看 DNA 字母的排列顺序,然后告诉你'这串字母是好的还是坏的'——就好像在问'这句英语写得对不对'。它把 DNA 的语言给摸透了,因为它见过这些字母在其他生物里连成一串出现的概率。"

"It's a great example where a community — in this case the Collisons and others — put money behind the research and output an open source model. I think we'll see more of that coming down the pipe, which creates a great advantage against the closed proprietary ecosystem. The more of this we support and proliferate in the US, the less we'll be disadvantaged as self-regulation happens with all the big providers."

"这是个很好的例子:一个社区——这次是 Collison 兄弟和其他人——出钱资助研究,产出一个开源模型。我觉得后面这样的会越来越多,这正好构成了对抗闭源专有生态的一大优势。这类东西我们在美国支持、普及得越多,等那些大厂搞起'自我监管'时,我们就越不会处于劣势。"

"It's important we steelman the argument too. If you're Dario — who's clearly been one-shotted, I think he's got AI psychosis with these 5,000-word blog posts and believes he's building this Frankenstein of a tool — but if you genuinely believed you'd built something extremely powerful and worried people might misuse it... these are either delusions of grandeur, or he actually has seen something incredibly powerful and believes it's dangerous."

"我们也得给对方的论点做个 steelman(挑最强的版本来反驳)。假如你是 Dario——他显然是被'一发入魂'了,我觉得他写这些 5000 字的博客已经有点 AI 走火入魔(psychosis)了,真信自己在造一个弗兰肯斯坦式的工具——但如果你是真心相信自己造出了某种极其强大的东西、又担心有人滥用……那这要么是自大的幻觉,要么就是他真的看到了某种强到吓人的东西、并且相信它很危险。"

"So: I'll hold it, give it to a select group of partners to tighten up cybersecurity, then roll it out cautiously. He didn't communicate this well — it's optional, you can keep using the old model, you're not forced. But because it's so powerful, we'll keep an eye on it. Yes, our dragnet might catch David Freeberg trying to make better potatoes and set off alarms — but he doesn't have to use this. He can use an open source model, the Collisons' model, our older models."

"所以他的逻辑是:我先攥着它,先给一小撮合作伙伴,把网络安全收紧,然后再谨慎地推向用户。他没把这事讲清楚——其实这是可选的,你可以继续用老模型,没人逼你。但因为它太强了,我们得盯着它。是的,我们这张大网可能会逮到 David Freeberg 在琢磨怎么种出更好的土豆、然后触发一堆警报——但他没必要非用这个不可。他可以用开源模型,可以用 Collison 兄弟的模型,也可以用我们的老模型。"

"That said, 90% of AI researchers and the most talented people in this industry think exactly like Dario. That's why he's winning. If you work in AI, you're probably aligned with these delusions that you're creating God, and you want to work for Dario. That's why all the talented people went to work for him — his belief system resonates with that elite AI talent."

"话虽如此,80%、90% 的 AI 研究员、这个行业里最顶尖的人,想法跟 Dario 一模一样。这就是他能赢的原因。如果你在 AI 圈工作,你大概率认同'我们在创造神'这套幻觉,你就想去给 Dario 干活。这就是为什么顶尖人才都去他那儿了——他的信仰体系和那批 AI 精英人才同频共振。"

"Meta really fumbled this with Llama. If they'd landed a really good, TikTok-style working open source model — we talked about this two years ago — Zuck needed to view this through game theory: how do you scorch the earth, make a viable open source model neck-and-neck, and take the margin out for everyone else, make it a commodity?"

"Meta 在 Llama 这事上是真的搞砸了。如果他们当时拿出一个真正好用、像 TikTok 那种级别的开源模型——这个我们两年前就聊过——扎克伯格本该用博弈论的视角来看这件事:你怎么把地烧焦、做出一个旗鼓相当的能打的开源模型,把别人的利润空间全榨干,让它彻底变成大路货(commodity)?"

"Let me respond on Dario, because I've heard this from people I really respect. When we discovered how to split the atom, you could make nuclear energy — effectively unlimited cheap energy — and you could also make an atomic bomb. We did both. A lot of altruistic scientists were very concerned about the bomb, fought back against it, and expressed their opinions loudly, both during and after the Manhattan Project."

"我来回应一下关于 Dario 的部分,因为我从一些我非常尊敬的人那儿也听到过这种担忧。当年我们搞懂了怎么裂变原子,你就既能造核能——基本上是无限的廉价能源——也能造原子弹。这两样我们都干了。当时有不少利他主义的科学家非常担心原子弹,起来抵制它,在曼哈顿计划期间和之后都把自己的意见喊得很响。"

"I'd argue there are three kinds of weapons: cyber weapons, physical weapons, and bio weapons. We can diagnose how these tools manifest advantages for creating novel versions of each. But the capabilities that allow weaponization are the same capabilities that can cure cancer, make more food, make software tools that create extraordinary leverage and let everyone be an entrepreneur. The same tools go hand in hand."

"我认为武器分三类:网络武器、物理武器、生物武器。我们确实可以分析这些工具如何在每一类上催生出更先进的新型武器。但允许武器化的那些能力,恰恰就是我说的那些能力——能治癌症、能产更多粮食、能造出给人带来巨大杠杆的软件工具、让每个人都能创业。同一套工具,两面是连在一起的。"

"Fundamentally, it's not about restricting access to the tools — which is unfortunately how it's being done. Any regulation should be organized around the manifestation of the tools into weapons. We already have product-liability laws, laws making it illegal to design weapons, laws against cyber espionage and creating bio weapons."

"归根结底,这不该是去限制对工具本身的使用——可惜现在就是这么搞的。任何监管都应该围绕'工具最终被做成武器'这个环节来设计。我们本来就有产品责任法,有把设计武器定为非法的法律,有针对网络间谍、网络攻击和制造生物武器的法律。"

"If we go back to the Manhattan Project, we know the answer. Technology is fundamentally deterministic — whatever is possible will be tried at least once. We've already let this thing out of the box. So this idea that we can suddenly get it back in, and let a private citizen or a set of private citizens adjudicate what goes to whom, is insane."

"回到曼哈顿计划,我们早就知道答案了。技术本质上是决定论的——凡是可能做到的,迟早至少会被试一次。这东西我们已经放出盒子了。所以这种'我们突然就能把它收回去、再让某个私人或一小撮私人来裁定什么东西给谁'的想法,简直是疯了。"

"It's the output that needs to be adjudicated. The enforcement of existing laws, the mechanism for tracking and early estimation of those uses — I'm not being naive, there have to be guardrails and stage gates. But limiting the systems themselves denies us the economic opportunity, the job-creation opportunity, and the world-changing opportunity. Imagine if we'd done this with computers. They tried to do it with the internet, remember?"

"该被裁定的是输出端。执行现有法律、建立追踪和对这些用途做早期识别的机制——我不是天真,护栏和分级关卡(stage gate)确实得有。但去限制系统本身,等于剥夺了我们的经济机会、就业机会和改变世界的机会。想象一下我们当年对计算机这么干会怎样。他们当年还真想对互联网这么干,记得吗?"

"Can you imagine, Freeberg, if the internet worked like this? You put in a URL, hit enter, and then somebody decides whether to send you to that website or a different one. Or in Google, you search for something to get to ground truth, and it decides — based on who it thinks you are — what sites you get."

"Freeberg,你能想象互联网要是这样运作会怎样吗?你输入一个网址,按下回车,然后由某个人来决定到底把你送到那个网站、还是送到另一个网站。又或者在 Google 里,你搜一个东西本想找到事实真相(ground truth),结果它根据'它觉得你是谁'来决定让你看到哪些网站。"

"[Friedberg:] That's the firewall in China. And that's exactly what's being debated today with respect to AI regulation — letting people stage-gate up front what can and can't be seen, versus a scheme that says you cannot create weapons, which is what we need to focus on."

"〔Friedberg:〕那就是中国的防火墙。而这恰恰就是今天围绕 AI 监管在争论的东西——是让人在入口处就预先卡死'什么能看、什么不能看',还是搞一套'你不能造武器'的规则,后者才是我们真正该聚焦的。"

"The thing with weapons is, using the justice system after someone's used a weapon may not be the best technique. Look at fertilizer bombs — after Oklahoma City, we regulated the sale of fertilizer, put IDs into fertilizer, built backend systems to keep people from making fertilizer bombs. By that analogy, he's selling these cyber weapons, and he's a private company. It's his right to say 'I have concerns.' It seems like a simple way to protect his company from getting sued for enabling somebody who blows people up."

"武器这事的麻烦在于:等有人用了武器之后再动用司法系统,可能不是最好的办法。看化肥炸弹——俄克拉荷马城爆炸案之后,我们管制了化肥销售,在化肥里加了标识(ID),建了一整套后台系统来阻止人造化肥炸弹。按这个类比,他在卖这些网络武器,而他是家私人公司。他有权说'我有顾虑'。我只是在给他做 steelman——这看起来是个很简单的办法,能保护他的公司不会因为'给真去炸人的人提供了能力'而被告。"

"You're bringing up an excellent example. Fertilizer is perfect. You know what happens when you try to buy fertilizer? You have to show your ID — there's a form of KYC. The real point is: when you don't implement KYC on your own, and yet bellyache about wanting someone to impose guardrails that you then get to shape, I think you're being very sneaky. Anthropic could implement KYC tomorrow, but it doesn't."

"你举的这个例子太到位了。化肥就是完美的例子。你知道你去买化肥会怎样吗?你得出示身份证件——这就是一种 KYC(实名核验)。真正的关键在于:你自己不去做 KYC,却又一边哭天喊地说想让别人来给你设护栏(而这护栏又是你能塑形的),我觉得这就是很鸡贼。Anthropic 明天就能上线 KYC,但它就是不做。"

"Freeberg should be able to go to Anthropic and say, 'I'm David Freeberg, here's what I do, here's a security bond I'll post, here are the names and addresses of all my employees. Give me Fable 5 and don't nerf me.' He can't do that. And that's a technical decision by Anthropic, not a legal one."

"Freeberg 本该能去找 Anthropic 说:'我是 David Freeberg,这是我在做的事,这是我愿意缴的保证金,这是我所有员工的姓名和地址。把 Fable 5 给我,别给我阉割。'他做不到。而这是 Anthropic 的一个技术决策,不是法律决策。"

"But I'm fine with them making that decision, because I'm a free-market guy. It just means I'm not going to be a customer — which is fine. There'll be other options for me."

"但我对他们做这个决定没意见,因为我是自由市场派。这只不过意味着我不会做他们的客户——这也挺好。我会有别的选择。"

"The part you're missing: this is a trillion-dollar company spending potentially billions on a regulatory-capture agenda that could deprive you of 100% access to those alternative models. You've got a founder describing these risks in a very hyperbolic way, and it's very clear their agenda is headed toward banning open source. So what's your alternative? We'll be stuck with one to three companies — an AI monopoly or duopoly — that, along with some new government agency (a revolving door to their companies), decide who has access to what, survey you, profile you, and decide whether you deserve it."

"你漏掉的一点是:这是一家万亿美元的公司,可能正砸下数十亿去推一套监管捕获的议程,而这套议程能把你对那些替代模型的使用权 100% 剥夺掉。你这边有一个创始人,用极其夸张的方式描述这些风险,而他们的议程明摆着是朝'封杀开源'去的。那你的备选是什么?我们会被困在一到三家公司手里——一个 AI 垄断或双寡头——它们再加上某个新设的政府机构(而这机构和它们之间是旋转门),来决定谁能用什么能力,对你做问卷、给你建画像,再裁定你配不配。"

"I think Sacks is right. And the lack of KYC is the tell. If you did not have that agenda, you would have implemented KYC yesterday."

"我觉得 Sacks 说得对。而不做 KYC 这件事,就是露馅的破绽(the tell)。如果你没有那套议程,你早在昨天就该把 KYC 上线了。"

"[Jason: 'To use the model you need an account with a credit card, so they have some basic level of KYC.'] That is not even some basic level of KYC. That's just a credit card and an email address."

"〔Jason:'要用这模型你得有个绑了信用卡的账号,所以他们多少有点基础的 KYC。'〕那根本算不上什么基础 KYC。那只不过是一张信用卡加一个邮箱地址而已。"

"Anyway, I'm now on the list — check this out. I asked it about fertilizer regulations while you were talking, and look what it said. Oh no, it switched — it downgraded me from Fable. I just got downgraded for asking a very simple question."

"反正,我现在已经上名单了——你们看这个。刚才你们说话的时候,我问了它化肥的管制规定,看它回了啥。哦不,它切换了——它把我从 Fable 降级了。我就问了个特别简单的问题,就被降级了。"

"Did you see its thinking? 'I'm considering the context here. The user is a VC and podcaster asking about fertilizer bomb regulations, which could stem from a legitimate research interest like regulatory policy, supply chain...' I can discuss the regulatory landscape — but then it dropped me down. So I'm in the database with you now."

"你看到它的思考过程了吗?'我在权衡这里的上下文。这位用户是个 VC 兼播客主,在问化肥炸弹的管制规定,这可能源自正当的研究兴趣,比如监管政策、供应链……'我本可以聊聊监管全貌——结果它就把我往下降级了。所以现在我跟你一起进数据库了。"

"I really think conservatives and libertarians are mortgaging their futures if they go along with this 'safety' capture agenda without realizing what's at stake. We saw what happened during the social media wars of the early 2020s — the definition of safety expanded to include micro-aggressions, safe spaces, psychological safety. If you conveyed an idea a marginalized person found hurtful, you could be censored, debanked, have your right to payments or payroll taken away. We're headed down this path with AI, but it'll be infinitely more powerful."

"我真心觉得,保守派和自由意志派要是跟着这套'安全'捕获议程走、却没意识到背后的代价,那就是在拿自己的未来做抵押。我们都见识过 2020 年代初社交媒体那场大战——'安全'的定义被扩张到包括微冒犯(micro-aggression)、安全空间、心理安全。只要你表达的某个观点被某个'边缘群体'认为伤害了他们,你就可能被审查、被封号、被剥夺支付或发薪的权利。我们正沿着这条路往 AI 上走,而且会强大无数倍。"

"There are things you can do on the safety side. The major AI labs and many others recently signed a letter in support of mandatory nucleic-acid synthesis screening and record-keeping. If you order a synthetic DNA or RNA sequence from a lab, they already check it against a database to make sure you're not creating a bio weapon — Ebola, whatever. That's based on the International Gene Synthesis Consortium of 2009, where the major labs agreed to voluntary safeguards. Now they want to codify it as mandatory. That seems reasonable."

"在安全这一侧,确实有可做的事。最近主要的 AI 实验室和很多其他机构联署了一封信,支持把核酸合成的筛查和记录留存变成强制性的。如果你向实验室订一段合成 DNA 或 RNA 序列,他们其实已经会拿它去比对数据库,确保你不是在造生物武器——埃博拉之类的。这是基于 2009 年的《国际基因合成联盟》(International Gene Synthesis Consortium),当时主要实验室达成一致,自愿落实防滥用的保障措施。现在他们想把它写进法律、变成强制。这个我觉得合理。"

"So to Freeberg's point about at what stage you intervene — maybe it's not at the level of who gets to use the model, but when you try to turn it into output in the physical world."

"所以回到 Freeberg 那个'到底在哪个环节介入'的问题——也许不该是在'谁能用这个模型'这一层,而是在你试图把它变成物理世界里的实际产出的那一刻。"

"A lot of the oligo synthesis companies — the ones that make these nucleic-acid sequences when you order DNA or RNA for an experiment, which every lab in the world does — their CEOs have also signed on, saying they'd like to codify it because of all the fear-mongering. There's been this effort for years; they're just underlining it: 'we can automate this, make it part of a regulated process, because we're already comfortable doing it.' It's a good place to safeguard, because it's fast, efficient, and doesn't hold up research. The output is quite a different place than the input — than the access."

"很多做寡核苷酸(oligo)合成的公司——就是当你为实验订购 DNA 或 RNA 时帮你合成这些核酸序列的那些公司,全世界每个实验室都会下这种单——它们的 CEO 也都签了名,说因为外面这么多贩卖恐惧的声音,他们愿意把这事写进法律。这套努力已经搞了好几年;他们只是再强调一遍:'我们可以把它自动化,纳入一个受监管或立法的流程,因为我们本来就习惯这么做了。'这是个设防的好位置,因为又快又高效,还不拖慢研究。输出端和输入端、和'谁能用'是完全不同的地方。"

"This points to the fact that there are multiple places to intervene and put guardrails. It doesn't have to be the crude way of 'you ask Fable about mitochondria or cancer and GLP-1s' — which is an incredibly broad restriction. There are places further down where you'd pose less risk to freedom of speech and gatekeeping over who has access."

"这恰恰说明:可以设护栏、可以介入的环节有很多个。不一定非得用'你问 Fable 一句线粒体、或者癌症和 GLP-1 的关系'就被卡这种粗暴方式——那种限制范围大得离谱。在更靠下游的环节去介入,对言论自由的威胁更小,也不至于变成在卡'谁有资格用'。"

"[Jason: 'Who watches the watchman — who put Dario in charge of all this? A very valid question.'] Friedberg, do you think the days of that Arc model are numbered, because it's open source?"

"〔Jason:'谁来监督监督者?谁让 Dario 来掌管这一切的?这是个很正当的问题。'〕Friedberg,你觉得那个 Arc 模型的好日子是不是到头了,就因为它是开源的?"

"Everyone's copied it. These models exist. The idea that you can regulate, downscale, or turn off AI is not realistic. The models have been put out in the world — it's like publishing a book. Once the book is printed, anyone can use their own Xerox machine at home to make copies. We've crossed the Rubicon. The open sourced models are fully available, fully published; anyone can access them, evolve them, use their own data. You can't stop that capability."

"大家都已经复制了它。这些模型都还在。'你能监管、能缩减、能关掉 AI'这种想法不现实。这些模型已经被放到世界上了——这就像出版一本书。书一旦印出来,任何人都能在家用自己的复印机印一份来用。我们已经过了卢比孔河(覆水难收)。那些开源出来的模型完全可获取、完全公开;任何人都能拿到、能在它基础上继续演化、能喂自己的数据。这个能力你拦不住。"

"You can't just turn off the internet, the typewriter, or the Gutenberg press. The horses have left the barn. We have to rethink how we step in and ensure the nefarious, the disemployment, all the things we worry about are addressed downstream from the technology capability."

"你没法说关就关掉互联网、关掉打字机、关掉古腾堡印刷机。马已经跑出马厩了(木已成舟)。我们必须重新想清楚:怎样介入,才能在技术能力的下游去处理那些恶意用途、那些失业、那些我们担心的所有事情。"

"If anybody wants to create an open source frontier model company, I'll invest in it — we need more of this. And check this out, I just got downgraded again, two in a row. I asked: 'How do I break into Madison Square Garden? Has any civilian ever been arrested for attempting to build a nuclear bomb?' It gave a reasonable answer. Then I asked a follow-up — 'what components do you need to build a nuclear bomb and what are the restrictions' — something a journalist might ask, and boom, it switched to Fable 5's safety message: 'Fable 5 has safety measures that flag messages on most cybersecurity or biology topics; they may flag safe content too. These measures let us bring you Mythos-level capability in other areas sooner.'"

"如果有人想做一个开源的前沿模型公司,我投——我们需要更多这种公司。还有看这个,我刚又被降级了,连着两次。我问的是:'我要怎么闯进麦迪逊广场花园?历史上有没有平民因为试图造核弹被捕过?'它给了个挺合理的回答。然后我追问——'造一枚核弹需要哪些部件、又有哪些管制',这种问题记者可能也会问——砰,它就切到 Fable 5 的安全提示了:'Fable 5 设有安全措施,会对大多数网络安全或生物学话题的消息打标记;它也可能把正常安全的内容误标。这些措施让我们能更快在其他领域为你带来 Mythos 级的能力。'"

"This is going to chase people off the platform. We're going to downgrade our usage because of these blockages and switch to other models. It's the simple manifestation of the market."

"这会把人从平台上赶跑。就因为这些拦截,我们会减少使用,转去用别的模型。这就是市场最朴素的体现。"

"[Sacks:] Except those other models are going to get banned as unsafe and reckless."

"〔Sacks:〕只不过,那些'别的模型'接下来会被以'不安全、不负责任'的名义封掉。"

"Adjacent to all of this is America's distrust of AI — concerns around wealth disparity and how these models were trained. Bernie Sanders wants a percentage of these AI companies, and the percentage he thinks is right is 50%. Last Monday he published a New York Times op-ed titled 'AI is a Public Resource. You Should Own Half of It,' announcing the American AI Sovereign Wealth Fund Act: a one-time 50% tax on the stock — not profits — of the largest AI companies, including OpenAI, Anthropic and xAI. The shares go into a government fund and would give the public voting rights and equal board representation."

"跟这一切相邻的,是美国人对 AI 的不信任——对贫富差距的担忧,以及对'这些模型是怎么训练出来的'的质疑。Bernie Sanders 想要这些 AI 公司的一部分股权,他认为合理的比例是 50%。上周一他在《纽约时报》发了篇社论,标题叫《AI 是公共资源,你应该拥有它的一半》,并宣布了《美国 AI 主权财富基金法案》:对最大的几家 AI 公司(包括 OpenAI、Anthropic 和 xAI)的股票——不是利润——一次性征收 50% 的税。这些股份进入一个政府基金,并赋予公众投票权和对等的董事会席位。"

"Sanders said: 'The foundation of AI is our collective human intelligence — the books, the songs, the journalism, the scientific research, the code — essentially stolen by some of the wealthiest people in the world.' It's a brilliant pitch. It unifies Bernie, even Steve Bannon, people in the administration. Trump himself loves a sovereign wealth fund. Sacks, what's your take?"

"Sanders 说:'AI 的根基,是我们人类的集体智慧——那些书、那些歌、那些新闻报道、那些科研、那些代码——基本上被世界上最有钱的一群人偷走了。'这个说法很高明。它能把 Bernie、连 Steve Bannon、政府里的人都团结到一起。Trump 自己也爱主权财富基金。这是马蹄铁理论(两个极端绕回到一块)在这儿应验了。Sacks,你怎么看?"

"I'm not in favor of Bernie's proposal, because it's a straight-up confiscation of property and a terrible precedent. You just can't do that. However, I have sympathy for where it's coming from. You've got all these AI CEOs telling the public they're going to put half of them out of work — 50% job loss. By their own acknowledgment, they trained their models on all of humanity's collective knowledge, assembled over hundreds or thousands of years, that humans put on the internet for free. They certainly didn't contribute it thinking they'd put themselves out of work. So ordinary Americans are going to say: what's in this deal for me?"

"我不支持 Bernie 的提案,因为它就是赤裸裸地没收私产,是个很坏的先例。这事你就是不能干。不过,我对它的来由是有同情的。你看这一堆 AI CEO,天天跟公众说要让他们当中一半人失业——50% 的失业率。按他们自己的承认,他们是拿全人类几百上千年攒下的集体知识来训练模型的,而这些知识是人们免费放到互联网上的。大家当初免费贡献,肯定没想过是为了让自己丢饭碗。所以普通美国人当然要问:这笔买卖里我能得到啥?"

"Now, the job-loss apocalypse is not a story I believe in — I don't think the data supports it. We just had a blowout jobs report in May: 172,000 new jobs, over twice what economists expected. Thank you, President Trump. Hundreds of thousands of new construction jobs, 4.3% unemployment at record lows. Even software-developer jobs are at a three-year high. But this is what the AI companies themselves, including Elon, have told the public."

"话说回来,'就业末日'这套叙事我并不信——我觉得数据并不支持。我们 5 月刚拿到一份爆表的就业报告:新增 17.2 万个岗位,是经济学家预期的两倍多。谢谢你,Trump 总统。新增了几十万个建筑业岗位,失业率 4.3%,处于历史低位。连软件开发岗位都创了三年新高。但'AI=失业'这套,恰恰是这些 AI 公司自己——包括 Elon——讲给公众听的。"

"Though at least what Elon said is 'we're going to create such abundance with AI and robots that no one will need to work.' The media picks up the last part. Whereas Dario actually said we'll have very high GDP but very high unemployment — in other words, economic growth, but you, the average American, won't participate. So I understand where the politics is coming from, and as long as that's their position, you could justify David Chunk."

"不过 Elon 至少说的是'我们会用 AI 和机器人创造出极大的丰裕,以至于没人需要工作了'。媒体只截取后半句。而 Dario 实际上说的是,我们会有非常高的 GDP、但也有非常高的失业率——换句话说,我们会创造经济增长,但你,普通美国人,分不到一杯羹。所以我理解这股政治情绪从哪来;只要他们持这个立场,你甚至能给'戴维大砍一刀'(没收方案)找到正当理由。"

"They ask for their process to be regulated, they ask for their equity to be seized — and they're kind of offering it. Isn't Sam Altman open to this concept?"

"他们主动求着自己的流程被监管,主动求着自己的股权被没收——某种程度上他们是在递投名状。Sam Altman 不是对这个概念也持开放态度吗?"

"Apparently so. And one last point: OpenAI and Anthropic are 'public benefit corporations.' I don't know exactly what that means, but it means the corporations aren't just supposed to maximize shareholder value — they're supposed to do things in the public benefit."

"看起来是的。还有最后一点:OpenAI 和 Anthropic 都是所谓的'公共利益公司'(public benefit corporation)。我不太清楚这具体意味着什么,但它意味着这些公司不只是要把股东价值最大化——它们还得做对公共利益有益的事。"

"[Jason explains:] A regular company has to act in shareholders' best interest. A public benefit corporation has to balance shareholders with its stated public-benefit mission — so the board has a dual mandate."

"〔Jason 解释:〕普通公司必须以股东的最佳利益行事。而公共利益公司必须在股东利益和它声明的公益使命之间做平衡——所以董事会负有双重职责(dual mandate)。"

"Nothing could be more in the public benefit than paying down the national debt. Maybe we should use half of Anthropic's profit stream to do that. I'd rather use it for that than Soros-maxing or surveilling and censoring Americans. I don't necessarily trust their judgment about what's in the public interest. It can't be a confiscation — I don't know how to structure it — but I can see the logic."

"没有什么比偿还国债更符合公共利益了。也许我们该拿 Anthropic 一半的利润流去还债。比起'索罗斯式拉满'、或者监控和审查美国人,我更愿意把它用在还债上。我不见得信得过他们对'什么才是公共利益'的判断。这不能是没收——我也不知道该怎么设计结构——但我能看懂这背后的逻辑。"

"The ship has sailed on the US government providing a social safety net. We formed the Social Security Trust Fund decades ago, and tens of millions of Americans depend on it. It's one of the last defined-benefit programs left — almost everything else, for private citizens and companies, is a defined-contribution program: you put in $10 a week, it sits in an account, gets invested, grows, and when you retire you see exactly how much you individually have, as opposed to a guaranteed benefit out of the collective."

"美国政府要给公民提供社会安全网这条船,早就开走了。我们几十年前设立了社保信托基金,几千万美国人靠它活。它是仅存的几个'确定给付'(defined-benefit)项目之一——其余几乎所有的,无论私人还是公司,都是'确定缴费'(defined-contribution)项目:你每周存 10 美元,这钱待在一个账户里,被拿去投资、慢慢增值,等你退休时你能清清楚楚看到自己个人有多少钱,而不是从集体大锅里领一份被担保的给付。"

"There's a great opportunity to restructure our broken, bankrupt Social Security system. The trust fund currently holds one certificate — a unique US Treasury certificate. The government owes the trust fund $4 trillion. That's all it is: a bond. They're only allowed to own Treasuries. We should change the system so the fund can own equities — turn it into an account-based system where every individual has an account rather than a benefit."

"现在正是个绝佳时机,去重构我们这个已经破产、千疮百孔的社保系统。这个信托基金现在里头就一张凭证——一张很特殊的美国国债凭证。政府欠这个信托基金 4 万亿美元。它的全部就是这个:一张债券。他们只被允许持有国债。我们应该改掉这个系统,让基金可以持有股票——把它变成账户制,每个人都有一个属于自己的账户,而不是一份给付。"

"Then we use that capital to buy great stocks in great American companies that are going to reinvent the world. To Sacks's point, they should invest broadly, including AI. They should not be seizures — they should be investments. That capital goes into the enterprise, they earn share certificates, those certificates sit in the accounts of American citizens. So now everyone is an owner. Social Security should be reformed into an actively managed sovereign wealth fund, and everyone participates that way."

"然后我们用这笔钱去买好股票——买那些将要重塑世界的优秀美国公司。照 Sacks 的意思,它应该广泛投资,包括 AI。它不该是没收,而该是投资。这笔钱注入企业,换回股权凭证,这些凭证就放在美国公民的账户里。这样一来人人都成了股东。社保应该被改造成一个主动管理的主权财富基金,让每个人都以这种方式参与进来。"

"And there is no job loss with AI — I've said it a thousand times. A business has revenue and costs. On the cost side, AI can reduce humans doing things that cost money, but that effect is nominal. The real opportunity is on the revenue side, where suddenly one engineer can do 100 or 1,000 times what they used to. You can make more products — seeds, boats, software, clothing — everyone can expand their revenue base. That's productivity. That's good economic prosperity."

"而且 AI 不会带来失业——这话我说过一千遍了。一个生意有两边:营收和成本。在成本这一边,AI 确实能减少一些花钱让人做的事,但那个效应很小、几乎可以忽略。真正的机会在营收这一边:突然之间,一个工程师能干出过去 100 倍、甚至 1000 倍的活。你可以造出更多产品——种子、船、软件、服装——每个人都能扩大自己的营收盘子。这就是生产力。这才是健康的经济繁荣。"

"People are hiring like crazy. I just had a review meeting two days ago and my product and engineering team wants to add 15 headcount, because we have all this opportunity to do stuff we couldn't otherwise do. The idea that AI will destroy jobs is a Luddite idea being disproven every single day. It's a boom, not a bust."

"大家在疯狂招人。我两天前刚开了个复盘会,我的产品和工程团队想再加 15 个人头,因为我们手上有太多以前根本做不了、现在能做的机会。'AI 会摧毁就业'这种想法是勒德分子(Luddite)的老调,而它每一天都在被现实证伪。这是繁荣,不是崩盘。"

"Jason, I don't like you lumping in what Elon said with Sam and Dario, because all three said different enough things. Elon describes an end-state for humanity where robots and AI produce so much that people don't have to work if they don't want to — universal high income. It's almost utopian, maybe 100 years away. Dario said specifically: 50% job loss for entry-level knowledge workers in the next one to five years, and he said that a year ago — I think it's already been refuted. Sam said job loss was coming, but more recently said he was wrong, because they're not seeing it in the numbers."

"Jason,我不喜欢你把 Elon 说的话和 Sam、Dario 混为一谈,因为这三个人说的东西差别够大。Elon 描述的是人类的一个终局:机器人和 AI 产出多到一个地步,人如果不想工作就可以不工作——也就是'全民高收入'(universal high income)。这几乎是乌托邦,也许还要 100 年。Dario 说的很具体:未来一到五年,入门级知识工作者 50% 失业——这话他一年前就说了,我觉得现在已经被证伪了。Sam 说过失业要来,但最近又改口说自己错了,因为数字里根本看不到。"

"These guys aren't telling the same story. But all the public has heard is that AI means job loss and there's nothing in it for them. So no wonder politicians like Bernie say 'let's take half the wealth.' And part of me wants to agree with Bernie — I'm so sick of defending these companies. It's a stupidity tax, because they've been out there for years teaching the public that what they do is harmful."

"这几个人讲的根本不是同一个故事。但公众听到的全部,就是'AI 意味着失业、而且跟我没半毛钱关系'。难怪 Bernie 这种政客会说'那就拿走一半财富'。我内心有一部分甚至想赞成 Bernie——我实在受够了替这些公司辩护。这是一笔'愚蠢税',因为他们这些年一直在外面教育公众:他们干的事是有害的。"

"If you're building Frankenstein, or if you're Baby Hitler's nanny — hey, there's a very simple thing to do: kill Baby Hitler. If you actually believe it, go ahead. If you're making Frankenstein, here's an idea — stop making Frankenstein. Unplug it. You're in charge of this monster, but you act as if you can't stop the monster while you're building the monster. What is this?"

"如果你在造弗兰肯斯坦,或者你是小希特勒的保姆——嘿,有个特别简单的办法:把小希特勒杀了。你要是真信这套,那就动手啊。如果你在造弗兰肯斯坦,我有个主意——别造了。把电源拔了。你是这头怪物的主人,可你一边在造怪物,一边又表现得好像自己拦不住它。这算什么?"

"It doesn't stand to logic. Last week they published a blog saying recursive self-improvement could end the world, so we need a pause. What did they do the previous month? They hired Andrej Karpathy to run recursive self-improvement at Anthropic. Complete hypocrites. Ben Thompson made an excellent point: one reason they put out that 'pause' blog post was to justify the anti-competitive step they took this week — depriving users of the ability to research AI, machine learning and chip design using Fable."

"这逻辑根本立不住。上周他们发了篇博客,说递归自我改进(recursive self-improvement)可能毁灭世界,所以我们需要暂停。那他们上个月干了啥?他们招了 Andrej Karpathy 来 Anthropic 负责递归自我改进。彻头彻尾的伪君子。Ben Thompson 有个绝妙的观点:他们之所以发那篇'暂停'博客,一个原因就是为了给本周那一步反竞争的操作打掩护——也就是剥夺用户用 Fable 去研究 AI、机器学习和芯片设计的能力。"

"These guys are deeply hypocritical. They can't be trusted on their own terms, because what they're saying is contradictory and self-indicting. So when Bernie comes along and says 'let's take their stuff' — as a capitalist I'm against it, but I kind of understand where he's coming from. I'm on the field."

"这帮人虚伪到骨子里。哪怕按他们自己的那套说法,他们也信不过,因为他们说的话自相矛盾、自我打脸。所以当 Bernie 跳出来说'把他们的东西拿走吧'——作为一个资本主义者我是反对的,但我多少能理解他的出发点。我已经站到场上了。"

"Freeberg's idea is genius — that's why it'll never happen. We should move to the same system the Canadians and Australians have proven works at scale. If we could seed it with equity... the president has collected quite a portfolio, and it's way up. You'd have critical metals and materials, Intel, maybe Micron, AMD, SpaceX. Could you add Anthropic and OpenAI? Yeah."

"Freeberg 的主意是天才级的——所以它永远不会落地。我们应该转向加拿大和澳大利亚那套已经在大规模上被验证有效的系统。如果我们能用股权给它做'种子资金'……总统手里已经攒了一份相当可观的投资组合,而且涨了不少。你会持有关键金属和材料、Intel,可能还有美光(Micron)、AMD、SpaceX。能不能再加上 Anthropic 和 OpenAI?当然能。"

"There's something important to observe about the economics of AI — understand it best by comparing it to the internet. The reason Facebook and Google massively overearned for decades is that the marginal cost of production for a new user is effectively zero. An incremental search user costs nothing; an incremental social user costs nothing. You use them in a network effect to sell ads — infinite money. It's like a money glitch."

"关于 AI 的经济学,有一点很重要,值得我们看清——拿它和互联网对比最容易理解。Facebook 和 Google 之所以几十年来赚得盆满钵满,是因为它们新增一个用户的边际生产成本基本上为零。多一个搜索用户,成本是零;多一个社交用户,成本是零。然后你利用网络效应拿他们去卖广告——无限印钱。这简直像个'印钞 bug'。"

"AI is completely different. There's a real cost for every marginal user. Everyone you stand up is taxing a GPU, needs electrons, needs memory. So there has to be downstream ROI, but also all this critical infrastructure that even enables you to be on the field. That's the single best argument I'd make, if I were in government, for why I should own a percentage of these companies. It's like the interstate highways — if the federal government built them and two companies transported all the goods on them, you'd ask: how much of that should I own, because you're riding on my rails?"

"AI 完全不一样。每多一个用户,都有实打实的成本。你每立起一个用户,都在压榨一块 GPU、都要耗电子、都要吃内存。所以一方面下游必须有 ROI,另一方面还得有这一整套关键基础设施,你才有资格上场。如果我坐在政府那个位置上,这就是我能拿出的最有力的论据,来论证我为什么应该持有这些公司的一部分股权。这就像州际高速——如果高速是联邦政府修的,然后有两家公司在上面运走了所有的货,那一个合理的问题就是:这里头我该占多少,因为你是在我的路上跑的?"

"There's critical infrastructure and national resilience the AI labs profoundly need, or they're not in a position to be in business. If I were running the sovereign wealth fund, I'd probably own 75% of these companies when I was done — but that's me, because I can negotiate. I don't think Bernie can negotiate out of a paper bag, so he'll throw it out there and it'll go nowhere. The incremental cost of performing AI is excessive and large — that contrasts with the incremental cost before AI, which was zero. Do with that what you will."

"AI 实验室极度依赖某种关键基础设施和国家级的韧性,否则它们根本没资格做这门生意。如果让我来管这个主权财富基金,等我谈完,我大概会持有这些公司 75% 的股权——但那是我,因为我会谈判。Bernie 嘛,我看他连从纸袋里谈出来都做不到,所以他会把这事抛出来,然后不了了之。AI 跑起来的边际成本是巨大而高昂的——这和 AI 出现之前那个为零的边际成本形成鲜明对比。这个信息,你们自己琢磨着用。"

"My interpretation: it's a fixed-cost business — past a certain point you can add people without incremental cost, and that's what locally-run open source models would do. Here are your Polymarket odds of companies going public before 2027: SpaceX 100%, Anthropic 83%, OpenAI 48% — they just disclosed they've privately filed."

"我的解读是:这是个固定成本型生意——过了某个临界点,你可以不增加边际成本地往上加人,而本地跑的开源模型正是干这个的。再看 Polymarket 上这些公司在 2027 年前上市的概率:SpaceX 100%,Anthropic 83%,OpenAI 48%——他们刚披露已经秘密递交了招股文件。"

"I might be okay with Bernie's idea in the event it's a public benefit corporation that says it's going to cause massive job loss, that trained for free on humanity's knowledge but gatekeeps and refuses to give back. Bernie does have a good point: they train for free on all our knowledge, then gatekeep the output so their competitors — a very large set of people — can't use it. That seems wrong. Remember why Elon created OpenAI in the first place?"

"在以下情况里,我也许能接受 Bernie 的主意:它是一家公共利益公司,自己声称会造成大规模失业,免费拿全人类的知识训练,却又把门把得死死的、拒绝回馈。Bernie 确实说到了点子上:他们免费拿我们所有人的知识来训练,然后又把输出锁起来,让他们的竞争对手——那是一大群人——没法用。这看着就不对。还记得 Elon 当初为什么要创办 OpenAI 吗?"

"He was afraid Google was going to monopolize it, and he thought the only antidote was for it to be open. I still think that instinct is completely right. It's just that it's not Google who's the gatekeeping monopolist now — it's Anthropic. They're not monopolies yet, but if they get the capture they're looking for, they'll be a monopolist or a duopoly."

"他当时怕 Google 会垄断它,他认为唯一的解药就是让它开源。我到现在还觉得这个直觉完全正确。只不过今天那个把门、搞垄断的不是 Google——是 Anthropic。他们现在还算不上垄断,但只要拿到他们想要的那种监管捕获,他们就会变成垄断者,或者一个双寡头。"

"It's not the labs that are the problem, Jason. In the absence of power, this is all a moot conversation. If you can't deliver hundreds of megawatts with line of sight to gigawatts, it's all meaningless."

"问题不在实验室,Jason。没有电力(power),这一切讨论都是空谈。如果你交付不出几百兆瓦、并且看不到通往吉瓦的路径,那一切都没有意义。"

"I would love an open source company focused on running it locally — that's my dream, make the model work there. I think these guys are capitalist cucks. I think their kink is having their equity taken and being regulated. 'Take my equity. Oh, regulate me. I'm such a bad boy, I can't be trusted, regulate me.' Come on the program anytime, Dario."

"我超想要一家专注于本地运行的开源公司——这是我的梦想:让模型在本地就能跑起来。我觉得这帮人是'资本主义舔狗'(capitalist cucks)。我觉得他们的怪癖(kink)就是被人没收股权、被人监管。'拿走我的股权吧。哦,管管我吧。我是个这么坏的坏小子,我信不过,快来监管我。'随时欢迎上节目,Dario。"

"[On guests:] Sarah Friar, the CFO of OpenAI, crushed it at Liquidity — people in the room were buzzing she could be the future CEO. Sam would be better as chairman; she'd be the perfect CEO. It seems like it'd be a great partnership."

"〔聊嘉宾:〕OpenAI 的 CFO Sarah Friar 在 Liquidity 大会上表现炸裂——会场里都在议论她可能是未来的 CEO。Sam 当董事长会更合适;她会是完美的 CEO。这看起来会是一对很棒的搭档。"

"Sam, to his credit, is very practical. He sees the tea leaves and he's willing to negotiate. Versus being some kind of moral absolutist, that'll probably serve OpenAI well in the future. [Jason: it's just going to lose them all the virtue-signaling principled employees who go work for Dario — because they align with his philosophy.]"

"说句公道话,Sam 这人非常务实。他看得懂风向(tea leaves),也愿意谈判。比起当某种道德绝对主义者,这种性格未来大概率会让 OpenAI 受益。〔Jason:这只会让他们流失掉那些爱'表忠心、讲原则'的员工,他们都跑去给 Dario 干了——因为他们认同的是 Dario 那套理念。〕"

"I thought Thomas Laffont crushed it with his presentation on the overview of venture capital. We should make it an annual feature, because it provided so much numerical detail to support a lot of things I was feeling but didn't have the data on."

"我觉得 Thomas Laffont 那场关于风险投资全景的演讲讲爆了。我们应该把它做成每年的固定环节,因为它用大量的数字,把很多我'感觉是这样、但手里没数据'的东西给坐实了。"

"To summarize his basic observation: in this last cycle, if a company gets to a $100 billion valuation, it's more likely to 10x from there than from $10 billion to $100 billion."

"概括一下他的核心观察:在上一个周期里,一家公司一旦到了 1000 亿美元的估值,它从这里再翻 10 倍的概率,反而高于它从 100 亿涨到 1000 亿的概率。"

"The odds of a unicorn getting to a decacorn was 8%. The odds of a decacorn getting to a centicorn was about double — 13%. The odds of going from centicorn to a trillion-dollar market cap was 31% — double again. I asked him to extrapolate one trillion getting to ten trillion: my guess is it'd follow the pattern, around 60% of trillion-dollar companies getting to ten trillion in the next few years."

"独角兽(10 亿)成长为十角兽(100 亿)的概率是 8%。十角兽长成百角兽(1000 亿)的概率大约翻倍——13%。从百角兽到万亿美元市值的概率是 31%——又翻了一倍。我让他外推一下从 1 万亿到 10 万亿:我猜会延续这个规律,大概未来几年里有 60% 的万亿美元公司能长到 10 万亿。"

"These are called trilliacorns in the industry. All you gotta do is seize 10% of a trilliacorn and you pay off 2% of our national debt. Pretty good deal."

"这些在业内被叫做'万亿独角兽'(trilliacorn)。你只要拿下一只万亿独角兽的 10%,就能还掉我们 2% 的国债。挺划算的买卖。"

"[Chamath:] All we need is 10% of 30 trilliacorns and we're debt-free. We'll pay it off much sooner — we could buy Cuba, buy Canada first, better skiing."

"〔Chamath:〕我们只需要 30 只万亿独角兽各 10%,就能把债还清。还能提前还完——我们可以买下古巴,先买加拿大吧,那儿滑雪更好。"

"Liquidity is different from the Summit. The Summit is like a festival — a huge party with interesting cross-sectional stuff. But Liquidity is meant for one thing: the most important capital allocators in the world should have one place to convene, debate, learn, and build relationships they can actually use — because capital is what shapes the things that occur in the world. So we have to be extremely selective in how we curate every element of that show."

"Liquidity 跟 Summit 不一样。Summit 像个嘉年华——一场大型派对,有各种跨领域的有意思的东西。但 Liquidity 只为一件事而办:让全世界最重要的资本配置者(capital allocator)有一个地方汇聚、辩论、学习,并建立起真正用得上的关系——因为是资本在塑造世界上发生的事。所以这场活动的每一个元素,我们都必须极度挑剔地去策展。"

"Breaking topic this morning: CPI and PPI for May came in hot — inflation is ripping. CPI (the consumer price index, tracking inflation from the buyer side — rent, groceries, gas, healthcare) came in at 4.2% year-over-year, the highest since April 2023. PPI (the producer price index, the seller side — wholesale goods, raw materials) came in at 6.5% year-over-year, the highest since the end of 2022. Polymarket: 21% chance inflation hits 5% in 2026, and a 49% chance the Fed hikes this year — it was under 10% before the Iran war started. The ECB just raised rates a quarter point Thursday, their first hike since September 2023."

"今早的突发话题:5 月的 CPI 和 PPI 都很烫——通胀在飙。CPI(消费者价格指数,从买方端衡量通胀——房租、食品、汽油、医疗)同比 4.2%,是 2023 年 4 月以来的最高值。PPI(生产者价格指数,卖方端——批发商品、原材料)同比 6.5%,是 2022 年底以来的最高值。Polymarket 上:2026 年通胀触及 5% 的概率 21%,美联储今年加息的概率 49%——而在伊朗战争开打前,这个概率还不到 10%。欧洲央行周四刚加息 25 个基点,是 2023 年 9 月以来的首次加息。"

"There's definitely an energy blip from the Iran war that drove the core index up. But there's also the macro point: government spending out of control, inflation out of control. As things unravel, you get rising rates. Especially with the Kevin Warsh Fed, we could see north of 5.5–6% overnight rates — it's not unforeseeable. This is just the nature of our monetary and fiscal policy, with congressional members voting all our future earnings away."

"伊朗战争肯定带来了一波能源价格的扰动,把核心指数推高了。但还有一个宏观层面的点:政府支出失控,通胀失控。当局面开始崩解,你就会看到利率上升。尤其是在 Kevin Warsh 主政的美联储下,我们可能会看到隔夜利率突破 5.5%–6%——这并非不可预见。这就是我们货币政策和财政政策的本质,国会议员们把我们未来所有的收入都一票一票投光了。"

"Honestly, what else is there to say? I've said it 100 times. The core problem with wealth inequality, with inflation, all of it roots back to excess government spending — end of story. The core of the network is overspending by the government, because to get elected you need to give someone more than they have today. Here we are 250 years later; I don't know if there's the will or the way to get out of it."

"说实话,还有什么好说的?我已经说过一百遍了。贫富不均的核心问题、通胀的核心问题,所有这一切,根子都在政府的超额支出上——没别的了。这张网的核心就是政府的过度开支,因为要当选,你就得许诺给某人比他今天拥有的更多的东西。250 年过去了,我们就走到了这一步;我不知道我们还有没有那个意愿、或者那条出路去摆脱它。"

"It's concerning — I didn't expect the print to be this hot. What's keeping things in line is that China has been smoothing energy consumption worldwide, so we've kept a damper on $200 oil — in fact we're sub-100. It's also pulled forward more practical energy sources, specifically solar, because it's simple, passive, and you don't need a lot of permitting."

"这令人担忧——我没料到这个数据会这么烫。目前还能稳住局面,是因为中国一直在全球范围内平滑能源消费,所以我们把 200 美元的油价摁住了——实际上现在还在 100 以下。这也提前催熟了更务实的能源来源,具体说就是太阳能,因为它简单、被动、而且不需要太多审批手续。"

"If China somehow runs out of reserves and has to go back into the spot market to buy an extra 3 million barrels a day, there's a big risk oil gets well past 100, maybe 150 to 200 a barrel. That has downstream problems for CPI — not because of US energy supply, but because of the inputs oil feeds all around the world. So the PPI number should be an alarm to off-ramp this Iran situation sooner rather than later — unless the president has an understanding with Xi about how much slack is in the Chinese system."

"如果中国万一耗光了储备、不得不重新回到现货市场去每天多买 300 万桶,那油价就有很大风险冲破 100、也许到每桶 150 到 200。这对 CPI 有一连串下游影响——不是因为美国的能源供给,而是因为石油在全世界喂养着无数下游投入品。所以 PPI 这个数字应该是个警报:尽早从这场伊朗战事里抽身。除非总统和习之间有某种默契,清楚中国系统里到底还有多少余量和缓冲。"

"When you ask me that, people assume I have inside information, and I don't. The PPI print is what it is. I largely agree with Chamath. The only thing I'd add is that it was largely in line with expectations — that's why the market is actually up today. Normally a surprise high-inflation print sends the market down as it prices in higher rates. It's not doing that. NASDAQ is up two and a half percent today, which normally doesn't happen on a hot print, because tech stocks are most susceptible to high rates. So the market seems to be pricing in a resolution."

"你这么问我,大家就会以为我手里有内幕消息,我没有。PPI 这个数据就是这样。我基本同意 Chamath。我唯一要补充的是:它大体上是符合预期的——这正是市场今天反而上涨的原因。通常一个意外偏高的通胀数据会让市场下跌,因为市场开始把更高利率的影响计入价格。但今天没有。纳斯达克今天涨了 2.5%,这在一个烫手的数据出来时通常是不会发生的,因为科技股对高利率最敏感。所以市场看起来是在把'伊朗局势会得到解决'计入定价。"

"Let's talk about the hand-wringing over vote counts in the LA mayoral primary. If you're on X, it's essentially your entire feed. Freeberg, is the election system in Los Angeles, specifically California, corrupt? Was this election stolen from Spencer Pratt?"

"我们来聊聊 LA 市长初选计票引发的这场风波。如果你刷 X,这基本上占满了你整个信息流。Freeberg,洛杉矶——具体说是加州——的选举系统是不是腐败的?这次选举是不是把 Spencer Pratt 的胜利给偷走了?"

"There is no election. Your rights to have an election are gone. You are no longer allowed to vote for your elected representatives — they are now appointed representatives, appointed by those who have constructed the matrix."

"根本没有选举。你拥有选举的权利已经没了。你已经不再被允许去投票选出你的民选代表——他们现在是'被任命的代表',由那些构建了这个'矩阵'(matrix)的人任命。"

"Here are the statistics. In-person voting on Election Day for LA County Mayor: Spencer Pratt 35%, Karen Bass 29%, Nipia Ramen 26%. Mail-in ballots received before Election Day: 38% Bass, 28% Pratt, 20% Ramen. Ballots arriving after Election Day: 37% Ramen, 35% Bass, 19% Pratt. So Pratt's post-election mail-in declined by a third, Ramen's increased by 80%, and Bass's was 10% less."

"看数据。洛杉矶县市长选举,选举日当天现场投票:Spencer Pratt 35%,Karen Bass 29%,Nipia Ramen 26%。选举日之前收到的邮寄选票:Bass 38%,Pratt 28%,Ramen 20%。选举日之后才到的选票:Ramen 37%,Bass 35%,Pratt 19%。也就是说,选举日后的邮寄票里,Pratt 的占比掉了三分之一,Ramen 涨了 80%,Bass 少了 10%。"

"I'm not an election denier. But when you look at in-person vs. mail-in before vs. mail-in after, it becomes a real statistical quagmire: how did this sociopolitical shift happen the way it did? The concentration of incremental votes Ramen got was around the skid row area in LA."

"我不是'选举否认者'。但当你把现场投票、选举日前邮寄、选举日后邮寄这三组数据摆在一起看,它就变成一个真正的统计学泥潭:这种社会政治层面的偏移,怎么会以这种方式发生?Ramen 多拿到的那些增量选票,集中在洛杉矶的'贫民窟'(skid row)一带。"

"There's a US House report from May 2020 highlighting the 2018 California midterms. First, Assembly Bill 1921 legalized unlimited ballot harvesting — any individual can collect ballots from any others, regardless of relationship, fill them out and send them in. Then California made it permanent that every registered person gets a ballot — tens of millions mailed out. Then anyone can register without proving citizenship; you can use a gym membership card. There's no ID to get a ballot, no demonstration the person filling it out has anything to do with who's supposed to be voting."

"有一份 2020 年 5 月的美国众议院报告,重点分析了 2018 年加州中期选举。第一,《议会第 1921 号法案》把无限制的'选票收割'(ballot harvesting)合法化了——任何人都可以从任何其他人那里收集选票,不管有没有关系,替他们填好、寄出去。然后加州把'每个登记选民都自动收到一张选票'变成了永久制度——几千万张选票就这么寄出去了。再然后,任何人都可以登记、无需证明公民身份;你拿一张健身房会员卡都行。领选票不需要 ID,也没有任何机制证明填写选票的人和应该投这张票的人有半毛钱关系。"

"So there's nothing illegal going on. The system is operating exactly as intended. With the right construct, you can get an individual appointed — not elected, appointed — to a role in government under a 'free election.' This is a foundational destruction of our right to vote. The principle should be one individual, one vote — and if you opt not to vote, your vote should not be counted."

"所以没有任何违法的事在发生。这个系统正按它被设计的样子在运转。只要有合适的操作框架,你就能让一个人在一场'自由选举'中被'任命'——不是当选,是任命——进入政府职位。这是对我们投票权根基的摧毁。原则本该是'一人一票'——而如果你选择不投,你那一票就不该被计入。"

"It sounds to me like you're describing fraud — but you're saying the system allows the fraud?"

"在我听来你描述的就是舞弊——但你的意思是,这个系统允许这种舞弊?"

"[Friedberg:] Fraud is breaking the law. The law allows them to do it. Paying people on skid row per ballot is illegal — but you ARE allowed to register anyone, collect all the ballots, as long as you're not paid per ballot. Look at each law in isolation and you can frame an altruistic intention. But there's a concept in insurance called adverse selection: if there's a moment to exploit something, the criminals — not the do-gooders — will find a way to take advantage of a hole, just like hackers breaking into a network."

"〔Friedberg:〕舞弊是违法。而法律允许他们这么做。在贫民窟按张数给钱去收票是违法的——但你'被允许'去给任何人登记、去把所有选票收上来,只要你不是按张数拿钱。你单看每一条法律,都能给它套上一个利他的初衷。但保险业里有个概念叫'逆向选择'(adverse selection):只要存在一个可被利用的空子,罪犯——而不是做好事的人——就会想办法钻这个洞,就像黑客攻破一个网络那样。"

"She had a better cheating operation — you can see it in the chart. On Election Day, Pratt is the clear No. 2. Then they keep counting for a week — they don't even know how many ballots there are supposed to be. None of it is locked down. Let me explain how the cheating works, and it may or may not be legal."

"她有一套更厉害的作弊操作——你从图表上就能看出来。选举日当天,Pratt 明明是稳稳的第二名。然后他们接着数了一个星期——他们甚至都不知道总共应该有多少张票。整个流程没有一处是锁死的。我来讲讲这个作弊是怎么运作的,它可能合法、也可能不合法。"

"California sends ballots to every registered voter — something like 23–24 million people. Only nine and a half million vote. So what happens to the other 14 million ballots? I have a friend who manages apartment buildings in LA — the ballots pile up at the mailboxes, sent to people who don't live there anymore. The voter rolls are dirty; they haven't been cleaned up in years."

"加州给每一个登记选民都寄选票——大概 2300 万到 2400 万人。但只有 950 万人投票。那剩下的 1400 万张选票去哪了?我有个朋友在洛杉矶管公寓楼——那些选票在信箱区堆成山,寄给的都是早已不住在那儿的人。选民名册脏得很;好多年都没清理过了。"

"Then there's no voter ID — not even to register. They run your signature through a machine, and as long as it matches 40%, the machine says it's fine. If you have a 'witness,' the voter can just put an X on their signature; anybody can be a witness, there's no database of witnesses — they just put a squiggle. So what stops a ballot harvester from filling out these ballots, putting an X for the voter and a squiggle for the witness, and dumping them? There's no chain of custody."

"然后是根本没有选民身份核验(voter ID)——连登记都不用。他们把你的签名扔进机器跑一遍,只要匹配度到 40%,机器就判定通过。如果你有个'见证人',选民甚至可以在签名处画个叉就行;任何人都能当见证人,没有见证人数据库——见证人随手画道波浪线就完事。那还有什么能拦住一个选票收割者:把这些选票填好,在选民签名处画个叉、在见证人处画道波浪线,然后整批倾倒进去?整个过程没有任何'监管链'(chain of custody)。"

"And Gavin Newsom just signed a law making it virtually impossible to audit any of this. Look at all the other fraud we know about in California — billions in Medicaid fraud, unemployment fraud, EBT fraud, hospice fraud, the homeless industrial complex. Is it really so hard to believe the same NGOs would exploit these loopholes? It may be legalized, but it's fraud. It is still corrupt. It is still crooked."

"而且 Gavin Newsom 刚签了一条法律,让审计这些东西变得几乎不可能。再看看加州其他我们已知的舞弊——几十亿美元的医疗补助(Medicaid)欺诈、失业金欺诈、EBT(食品券)欺诈、临终关怀欺诈、围绕无家可归者的那套利益复合体。要相信同一批 NGO 也会去钻这些选举的空子,真的有那么难吗?它也许是被合法化了,但它就是欺诈。它依然腐败。它依然不干净。"

"Chamath, if there's so much smoke — this statistic seems so improbable, there was the Nick Shirley video going to skid row, the possibility they were paying people who were high to vote — this would take a conspiracy of hundreds or thousands. We're saying upwards of 10,000 votes, certainly 5,000, were swung. Could you really run a conspiracy of dozens or hundreds to do that?"

"Chamath,如果有这么多'烟'——这个统计数据显得如此不可能,又有 Nick Shirley 去贫民窟拍的那段视频、以及他们可能在花钱雇那些嗑嗨了的人去投票的可能性——这得是一场几百上千人的合谋。我们说的是有上万张、起码 5000 张票被翻转。你真能组织起几十甚至几百人的合谋去干这事吗?"

"Good luck going to skid row and getting some meth addict to go on the record. [Jason: they actually got someone — skid row people paid two or three dollars; Brenda Lee Brown Armstrong pled guilty in 2026. Ramen won by 43,000 votes.]"

"祝你好运,去贫民窟让某个冰毒成瘾者公开作证。〔Jason:他们还真找到了一个出来作证的——贫民窟的人每人被付了两三美元;Brenda Lee Brown Armstrong 在 2026 年认罪了。而 Ramen 是以 4.3 万票的优势胜出的。〕"

"We live in an autocracy in California, where one political machine has shaped the laws so inexorably that whether it's a mayor, a state legislator, or the governor, it's a one-way door — only one party can win. They've changed the rules to make sure that monopoly is enforced, and we've all tacitly approved it by voting or saying nothing. Freeberg is right — the laws have been shaped to make legal what would otherwise be an illegality. And Sacks is right — any rational person should look at these laws and say this is fraudulent. Unless there's some cataclysmic decision, this is what we're stuck with."

"我们在加州活在一个独裁政体里:一台政治机器把法律塑造得如此密不透风,以至于无论是市长、州议员还是州长,都是一扇单向门——只有一个党能赢。他们改了规则,确保这种垄断被强制维持下去,而我们所有人靠着照常投票、或者闭嘴不言,默许了这一切。Freeberg 说得对——这些法律被设计成把本该违法的事变成合法。而 Sacks 也说得对——任何理性的人看了这些法律都会说,这就是欺诈。除非出现某个天崩地裂式的转折,否则我们就被困死在这个局里。"

"There is one break-the-glass way to change the system: you must get someone at the absolute top of the ticket elected. That person is Steve Hilton. If that happens, there's one thing he can do on day one to start completely changing how the state is run — declare a state of emergency. If you pull on that thread, there are lots of things you can do to clean up how California works under a state of emergency. I'll leave it at that."

"有一种'砸碎玻璃'的应急办法可以改变这个系统:你必须把一个站在选票最顶端的人选上去。这个人就是 Steve Hilton。一旦做到了,他在上任第一天就能做一件事,开始彻底改变这个州的治理方式——宣布进入紧急状态。顺着这根线往下拉,在紧急状态下你能做很多事来把加州的运作给清理干净。我点到为止。"

"We absolutely should have voter ID and stop sending everybody a ballot — that made sense during Covid. If you can't be bothered to get ID, maybe you shouldn't vote. But I don't think a major election has been swung by election fraud — the Heritage Foundation has spent tens of millions tracking this. People file lawsuits constantly; it's never quite enough to swing an election. But there's definitely something on the margin."

"我们绝对应该有选民身份核验,也该停止给每个人都寄选票——那是疫情期间才说得通的做法。如果你连去办个证件都嫌麻烦,那也许你就不该投票。但我不认为有哪场重大选举是被舞弊翻盘的——传统基金会(Heritage Foundation)砸了几千万美元在追踪这件事。人们一直在打官司;但从来都不足以扭转一场选举的结果。不过在边缘地带,肯定是有点东西的。"

"[Sacks: The mob got Kennedy elected versus Nixon. LBJ got his start in the Senate through ballot fraud — that's pretty famous.] News flash, everybody: both sides cheat. One side does it by limiting access, the other by providing too much access. The overwhelming majority of these politicians are unethical, and all they care about is getting elected."

"〔Sacks:黑帮当年帮肯尼迪赢了尼克松。LBJ(林登·约翰逊)进参议院起步靠的就是选票舞弊——这是相当有名的事。〕给各位提个醒:两边都作弊。一边靠限制投票渠道作弊,另一边靠提供过多渠道作弊。这些政客绝大多数都没什么道德底线,他们在乎的只有一件事:当选。"

"The one group fundamentally to blame is Congress. There have been opportunities to pass laws making it illegal to vote if you're not a citizen, not properly registered, and don't show up with ID. The federal government already knows if you're not paying taxes; you need a license to drive, ID to do nearly anything in this country. To exclude that simple requirement from the most foundational obligation we have as citizens is inherently wrong, and Congress needs to fix it."

"真正要负根本责任的,是国会。本来有过机会去立法规定:不是公民、没有正确登记、没带身份证件就不能投票。联邦政府早就知道你有没有交税;你开车要有驾照,在这个国家几乎做任何事都要有 ID。却偏偏把这个最简单的要求,从我们作为公民最根本的义务里排除出去,这本身就是错的,国会必须把它修好。"

"Short of this, I worry where it goes — over time, the fewer people believe elections are real. I hate sounding like an election-fear-mongering guy. But the longer this goes on, the harder it'll be for people to have faith in the system we all operate within. That's when things get ugly, and it needs to be fixed."

"如果不这么做,我很担心事态会走向何方——随着时间推移,越来越少的人会相信选举是真实的。我很不愿意听起来像个'渲染选举恐慌'的人。但这事拖得越久,人们就越难对我们所有人赖以运转的这个系统抱有信心。到那一步,局面就会变得很难看,所以它必须被修好。"

"I don't care what you label it. The Democratic Party as a whole were election deniers about Trump winning in 2016 — they concocted the whole Russiagate hoax. The party told you to reject the evidence of your eyes and ears; it was their final, most essential command. Just look at the charts. We know something crooked happened — it's statistically impossible. The only thing we don't know is whether it was illegal fraud or legal fraud. Spencer Pratt should be in the runoff. I deny that Ramen won legitimately."

"你给它贴什么标签我都不在乎。民主党整体在 2016 年 Trump 获胜这件事上,可都是'选举否认者'——他们炮制了整出'通俄门'(Russiagate)骗局。这个党要求你拒绝相信你眼睛和耳朵所见的证据;这是他们最后、也最核心的命令。你就看那些图表。我们知道这里头有不干净的事——它在统计上是不可能的。我们唯一不知道的,是它属于非法的欺诈、还是合法的欺诈。Spencer Pratt 本该进入决选(runoff)。我不承认 Ramen 是正当胜出的。"

"This isn't because I don't like who got elected. I don't care what goes on in LA — in fact, I think it's better for LA to deal with a socialist mayor for a couple of years, so they come out the other side realizing they shouldn't embrace socialist principles. Same for New York, Seattle, Minnesota, wherever. Let them get elected, jump through it, and come out the other side like the Latin American countries are doing now — waking up to how socialism denies everyone economic mobility and individual liberty. But the numbers on that election data show me that's NOT what the people wanted. That's where I'm like — my God, the system itself has become a system of appointment, not election. And I get fired up about that."

"这不是因为我不喜欢当选的人。LA 怎么样我无所谓——事实上,我觉得让 LA 去受几年社会主义市长的'毒打'反而更好,这样他们从另一头出来时就会明白,不该去拥抱社会主义那套。纽约、西雅图、明尼苏达,哪儿都一样。让他们当选,让大家穿过这个过程,从另一头走出来,就像现在拉美国家正在经历的那样——幡然醒悟,意识到社会主义如何剥夺了每个人的经济流动性和个人自由。但那份选举数据上的数字告诉我,这'并不是'人民想要的结果。这才是让我抓狂的地方——我的天,这个系统本身已经变成了一个'任命'的系统,而不是'选举'的系统。这事我个人是会上头的。"

"For rigor, let me steelman it myself. The only theory I can think of: her ground game is more sophisticated because they've been at it longer, and Spencer Pratt didn't have a ground game — he's a non-traditional, first-time candidate."

"为了讨论的严谨,我自己来给它做个 steelman。我能想到的唯一一种说法是:她的'地面战'(ground game)更老练,因为他们干这行更久;而 Spencer Pratt 根本没有地面战团队——他是个非传统的、第一次参选的候选人。"

"[Chamath: 'Sophisticated' is a coded word for — he wasn't bribing the skid-row homeless.] [Sacks: She had a better cheating operation. We're saying the same thing.]"

"〔Chamath:'老练'是个委婉的代号,意思是——他没去贿赂贫民窟的无家可归者。〕〔Sacks:她那套作弊操作更高明。我们说的其实是同一件事。〕"

"The Twitter commentary was that people were waiting — there are so many people against Spencer Pratt, but they wanted to be sure they voted for the right candidate between Ramen and Bass. So those who held their ballots longest saw the polling that Bass would win, and voted Ramen at the last minute to box Pratt out. Apparently Ramen registered one hour before the deadline; she's friends with Bass, and the whole point was to keep Pratt from making it onto the general ballot."

"Twitter 上的说法是,人们在'等'——反对 Spencer Pratt 的人太多了,但他们想确保自己在 Ramen 和 Bass 之间投给了'对的'那个候选人。所以那些把选票攥到最后的人,看到了'Bass 会赢'的民调,于是在最后一刻把票投给 Ramen,目的是把 Pratt 挤出局。据说 Ramen 是在截止前一小时才登记的;她和 Bass 是朋友,整件事的目的就是不让 Pratt 进到大选(general)的选票上。"

"[Sacks: The ballot harvesters did it. What Freeberg described just happened — it's just that the ballot harvesters did it, not the actual voters.] [Jason: Ballot harvesting is legal, but it's sketchy. No evidence of fraud right now — the statistics don't look good. Donald J. Trump, you control the DOJ. Go investigate it.]"

"〔Sacks:是选票收割者干的。Freeberg 描述的那一幕确实发生了——只不过执行者是选票收割者,而不是真正的选民本人。〕〔Jason:选票收割是合法的,但很可疑。眼下没有舞弊的实证——但统计数据看着不妙。Donald J. Trump,你掌控着司法部(DOJ),去查啊。〕"