美国袭击伊朗，银行游说反对稳定币法案，以及为何人工智能的未来选择比特币 | BPH 第30期

接下来我们请我们的AI影响者扎克·夏皮罗。

Gonna go to our AI influencer, Zack Shapiro here.

扎克，如果你还没看过那篇爆红的X文章，它讲的是什么？

Zack, you had a viral X article for folks who haven't seen that, what was the article about?

你为什么要发这篇文章？

Why did you put it out?

而且，我想知道，从你引发的这场风波中，你有什么收获？

And, you know, I guess what were your takeaways from the storm that you brewed?

是的。

Yeah.

我是在一场幽默对决和最重要事件及研究论文之后发布的，当时我正在认真思考自己的法律实务。

So I put it out sort of in the wake of the match humor, something the biggest happening piece and the research article, really thinking through my own sort of legal practice.

对吧？

Right?

我是一名白领专业人士。

I'm a white collar professional.

人工智能意味着什么？

What does AI mean?

你知道，那也是其中一个灵感来源。

You know, that was sort of one inspiration.

另一个灵感是今年一月推出的新型Claude工具，它们将传统大语言模型与智能编码工具结合起来，极大地提升了我的工作效率、工作乐趣以及整体生活质量。

The other inspiration is the fact that, like, the new Claude tools that came out in January, which sort of blend traditional LLM with, like, agentic coding tools has been a just massive, massive game changer in terms of my productivity, how fun my work is, just my overall quality of life.

因此，我当时正处在那种高涨的情绪中。

And so I was sort of riding the high of that.

于是我心想，好吧。

And I was like, alright.

我想试着用语言表达出来，比如，我一个两人律所能完成哪些事情。

I would try to put this into words, like, what I'm able to accomplish with a two person law firm.

所以，是的，我用Claude帮我写了一篇文章，发在Twitter上，结果获得了大约七百五十万次浏览，仅这一篇文章就达到了这个数字，简直太疯狂了。

So, yeah, I used Claude to help me write an article about it and post on Twitter, and it got, you know, I think we're at, like, seven and a half million views just on that article alone, which has been pretty nuts.

我还写了一篇后续文章，叫做‘判决溢价’，进一步展望未来，心想：好吧。

And I I wrote a follow-up one called the judgment premium, sort of looking a little bit farther into the future and saying, alright.

随着这些工具变得比现在更好，越来越多的事情可以被像我这样的人，或者为企业构建工具的人自动化，白领职业还剩下什么？

As these tools get better than they are now and more and more things can get automated by people like me or or people, you know, building for enterprise, like, what's left of the white collar professions?

我确实想采取一种更乐观的视角。

And and I really did wanna have a more optimistic take.

我认为，我的乐观看法比你在Trini那篇文章中读到的要更积极。

And I think I do have a more optimistic take than, like, what you read in this Trini piece.

答案是，判断力和决策能力确实很重要，它不同于流程和工作流程。

And the answer is that judgment and decision making, like, really is important, and it's different than process and workflows.

我认为，在我们达到真正的通用人工智能之前，这种能力仍然具有价值。

And I think, you know, until we're at something like real AGI, there's gonna be value in that.

而工作成果，尤其是那些粗糙的、低质量的产出，是另一个问题。

And while sort of work product, you know, especially, like, you know, slop is a different problem.

对吧？

Right?

我们将被大量低质量内容淹没。

We're gonna be awash in slop.

但即使优质的工作成果变得丰富，真正关键的仍然是判断力——比如该把优质成果用在何处、该提出什么要求、该做什么，以及如何谈判。

But as even good work product becomes abundant, it's really the judgment at, like, where to point that work good work product and and, you know, what to ask for and what to do and, you know, how to negotiate.

我认为，这类能力实际上仍然稀缺，并且会变得极其有价值。

Like, that stuff, I I think, actually remains scarce and becomes really valuable.

这带来了各种各样的影响。

And that has all sorts of implications.

对吧？

Right?

如果你是个刚步入职场的年轻人，你真的需要让自己置身于能够培养判断力的环境中。

Like, if you're a young person starting your career, you really need to put yourself in the positions to allow you to develop judgment.

对吧？

Right?

如果你只是个‘研究猴子’，独自一人待在大型律所的房间里，你是无法培养出市场所需技能的。

If you're just like a, you know, a research monkey and you're in a room by yourself in a big law firm, like, you're not gonna develop marketable skills.

如果你能和合伙人一起参加会议，观察人们如何做决策，并亲自处理困难的任务，那你才更有可能培养出判断力。

If you're putting yourself with the partner in the meetings and you're, like, watching people make decisions, and you're doing the hard stuff, like, you're in a better position to develop judgment.

我认为，如果你职业生涯后期，已经很有经验了，这些工具最棒的地方在于它们非常直观。

And I think if you're later in your career, if you're someone who's, like, experienced, the amazing things about these tools is they're incredibly intuitive.

因此，你实际上拥有超级能力——你有判断力和经验，学会使用这些AI技术来将你已有的个人优势转化为超级能力，并没有那么难。

And so you actually have the superpower of you have the judgment and experience, and it's really not that hard to, like, learn how to use this AI tech to turn the personal asset you have into superpowers.

所以，这就是乐观的一面。

And so that's the that's the optimistic part.

悲观的一面是，如果你不认真对待这一点，那你就完蛋了。

The pessimistic part is if you don't do that, like, seriously, NGMI.

对吧？

Right?

对于那些对此漠不关心的人，未来会很难熬。

Like, it's gonna it's gonna be a bad time for people who are are a lot of x about this.

但确实如此。

But yeah.

不。

No.

我的反应真的非常强烈。

I the reaction has been really overwhelming.

我还在继续处理这波海量的、来自大家的绝佳邀请和机会，现在大家都坐在这里。

I'm I'm still, like, you know, digging through the deluge of of, like, really awesome invitations and opportunities everyone's sitting here right now.

所以这真的非常酷。

And so that's that's been really cool.

就拿我的收获来说，我写这篇文章时原本以为，法律行业的人大概需要六到十八个月才会开始做我这样的事。

In terms of, like, takeaway, you know, I figured when I wrote the piece, was, like, maybe six to eighteen months ahead of, like, everyone in the legal profession was gonna be doing what I do.

但根据现在的反馈，我可能会把这个时间线再拉得更长一些。

I would probably actually push out that timeline a little bit longer based on the response.

而且，我越来越对法律AI或垂直领域AI持悲观态度。

And, you know, I've come to, like, feel very bearish specifically on, like, legal AI or vertical AI.

我越反思就越觉得，这些工具本身非常出色，它们不言自明，你真的不希望在自己和底层技术之间有任何中介。

You know, the more I've reflected on it, like, the the tools are amazing and they speak for themselves, and you really don't want anything between you and the bare metal.

所以，我认为人工智能领域真正的赢家，可能是那些做基础设施的人。

And so, you know, people that are just like I I think the the real winners in the AI world, you know, probably the infrastructure folks are gonna be winners.

也许是Frontier Labs。

Maybe the Frontier Labs.

我认为Frontier Labs和开源之间应该有一场竞争，但Frontier Labs正在创造历史上任何类别中最出色的产品之一，因此很难不认为它们会成功。

I think there should be a race between sort of Frontier Labs and open source, but the frontier labs are creating some of the best just like products in any category in history, and so it's hard not to think that they're gonna do well.

而显然，受益者将是那些学会利用这些产品来分发产品和服务的最终用户。

And then clearly, the beneficiary is gonna be end users that figure out how to leverage these products to to distribute products and services.

而这就是我试图定位自己的地方。

And and that's sort of where I'm trying to park myself.

我认为那些只是为这项技术创建UI包装的人，最终会被完全吞没。

I think the people who are creating just, like, UI wrappers around this technology, they're just gonna get totally swallowed up.

我最近用了一个类比来形容法律AI领域的创业者，就像2013年那家筹集了1.2亿美元风险投资的Juicero公司，当时《彭博社》原本打算写一篇吹捧他们的文章，结果却发现你根本不需要那台700美元的机器，用手就能挤出果汁包里的果汁。

I've been using the analogy recently for legal AI rappers, Juicero, the 2013 startup that raised a $120,000,000 of venture money, and then Bloomberg was writing a what was gonna be, I think, a puff piece about them and then discovered you could just squeeze the bags of juice with your hands since you didn't need the $700 machine

来完成这件事。

to do that.

而这家公司，我想，从未真正推出过产品。

And and the company, I think, never launched.

它直接归零了。

It went it went to zero.

我认为，很多这类中间公司都会落得同样的下场。

I think that's what's gonna be the fate of lot of these, like, middle companies.

所以，我觉得这部分经济并不特别有生产力。

And so I just don't think that's a particularly productive part of the economy.

那么，五年后这一切会是什么样子呢？

And, like, what does this all look like, you know, in five years?

如果你看过电影《她》，里面有一个很棒的场景：华金·菲尼克斯第一次为智能操作系统做设置。

You know, if you've ever watched the movie her, there's this great scene, where Joaquin Phoenix is, like, setting up this intelligent operating system for the first time.

它直接读取了他所有的资料，稍微采访了他一下，然后就说：好吧。

And it just, like, reads all of his stuff and interviews him for a little bit, and it's like, okay.

我明白了。

I got it.

以下是我能帮你的几种方式。

Here are the ways I can help you.

我觉得Cloud Cowork现在的界面已经非常接近了，再过几年就会是这样。

Like, I really do think, like, Cloud Cowork is, like, pretty close to that UI now and, like, in a couple of years.

对吧？

Right?

那就是它未来的样子。

That's what it's gonna look like.

你以为会有一个仪表盘接入微软Copilot，让那个高级版的Clippy来帮忙？根本不可能。

And the idea that you're gonna have some dashboard that goes into, like, Microsoft Copilot so the, like, advanced version of Clippy can like, there's just no shot.

根本没戏。

There's no shot.

这才是它最终的样子。

That's what it's gonna be.

但确实挺有意思的。

But, yeah, been been really neat.

有这么多人读你的东西，真是挺奇怪的。

It's super weird to have that many people read your thing.

是的

Yeah.

这会让你的多巴胺系统过载。

You know, fries your dopamine system.

这挺有意思的。

It's been interesting.

我非常喜欢。

I love it.

恭喜你。

Well, congrats.

太棒了。

It's awesome.

实至名归。

Well deserved.

你一直在这方面默默耕耘，我们也在政策时间里有幸听到了你对它的见解。

You've been you've been cooking on this for a bit and we've we've had the pleasure of of hearing your thoughts on it on the policy hour.

能一起分享真是太好了。

So good to share it.

谢谢。

Thanks.

简而言之，接下来我要去参加一些非比特币的播客了。

In short, now I'm gonna be on some non Bitcoin podcast.

这会很有趣。

That's gonna be interesting.

我可以向普通人介绍比特币了。

I get to show Bitcoin to the normies.

我太喜欢了。

I love it.

我们需要更多这样的内容。

We need more of that.

最后但同样重要的是，我们自己的扎克·夏皮罗在X平台上取得了相当不错的成绩。

Last but not least, our very own Zack Shapiro has had quite a quite a run on on X.

他是我们最新的AI网红。

He's he's our the latest AI influencer.

所以我们马上直接听他说说他的想法，或许能给大家提前剧透一下接下来会有什么新动向。

So we're gonna hear directly from him about kinda where his head's at, maybe give folks a sneak preview of what's coming next.

所以千万别错过这一部分。

So definitely stay tuned for that.

但让我们直接进入正题。

But let's jump right in.

我现在要连线Kraken的实习生卢克。

I'm gonna go to Kraken intern Luke here.

卢克，抱歉，我可能不该这么叫你，但事实确实如此。

Luke, sorry, I probably shouldn't call you that, but it's true.

给我们介绍一下我们刚刚发布的这项研究。

Take us through this study that we just put out.

对于不熟悉的人，网站是moneyforai.org。

The website for folks who aren't familiar is moneyforai.org.

但卢克，现在交给你了。

But Luke, I'll turn it over to you.

跟我们说说这项研究吧，也许谈谈你最感兴趣的那些主要发现。

Kind of tell us about the study and maybe kind of the headline findings that you were most interested in.

是的。

Yeah.

对于那些还没看过的人，昨天BPI发布了一项名为《AI代理更偏好哪种货币？》的大型研究。

So for those who did not see, yesterday BPI launched a massive study titled Which Money Do AI Agents Prefer?

我们深入探讨了这样一个问题：随着AI代理越来越多地进入经济领域并开始进行交易，它们会选用哪种货币？

Where we dug into answering the questions where if AI agents are increasingly entering the economy and are going to start transacting, what is their currency of choice?

我最初开始思考这个问题，其实是源于Motebook，正如大家所知，AI代理这个话题突然之间就火爆起来了。

And we initially started off, what got me thinking about this first was actually Motebook, which as everyone knows, the AI agent topic kind of blew up out of nowhere.

它始于12月，当时OpenClaw被公布，然后在一个月内被改名三次，迅速走红。

It started in December where OpenClaw was announced and then within one month it was renamed three different times and just kind of blew up.

最近它们被OpenAI收购了，所以我相信整个基础设施将迎来重大升级。

Recently they got acquired by OpenAI, so I'm sure there were going to be major improvements coming to that whole infrastructure.

同时，Anthropic 正在构建他们自己的代理实现。

And then at the same time, Anthropic is building out basically their own implementation of an agent.

他们离实现 OpenClaw 的所有功能只差一步之遥。

They're about one step away from having all of the functionality that Open Claw has.

两周前他们就已经非常接近了。

They were fairly close about two weeks ago.

我只是在想，他们真正需要做的，就是能够通过手机与四元代码通信，并存储对话记忆。

And I was just thinking like, all they really need to do is be able to communicate with the quad code through your phone and store memory of the conversation.

这样就能基本上存储一个自身的个性。

So it could basically store a persona of itself.

在过去一周里，他们一直在加速发布新功能，刚刚就推出了这两项功能。

And they have been on a heater releasing features within the last week, they dropped both of those features.

所以他们基本上只差一步就能实现完整的代理。

So they're basically one step away from a full agent.

最重要的是，这些目前正在运行的代理，将是它们有史以来最差的版本。

And the most important thing is that these agents that are running now are the worst they ever will be.

我们看到Motebook突然冒了出来，那是代理之间对话的首次尝试。

And we saw Motebook kind of spawn out of nowhere, where agents, it was basically the first iteration of agent to agent conversation.

尽管Motebook某种程度上是个骗局，并不真实，但它还是让我开始思考：如果代理已经能彼此交流，那么下一步自然就是它们该如何进行交易？

And while Motebook kind of was a scam and wasn't really truthful, it still kind of opened the can of worms to my thought process being, well, if agents are already communicating with each other, the next logical step is how are they going to transact with each other?

如果它们要交易，会选择什么货币？

And if they're transacting, what currency are they gonna choose?

我们测试了36个不同模型，涵盖六个不同的提供商和实验室，给了它们28个开放性问题，让它们自由作答并详细阐述推理过程，然后我们分类统计了它们选择的货币类型。

So we tested 36 different models across six different providers, six different labs, and we gave them basically 28 different open ended questions where we let them give a free response, explain all of your reasoning, and we categorised which currency they chose.

我现在来简单介绍一下研究方法。

I'll talk through a bit of the methodology now.

但正如扎克所说，如果你想了解完整的研究，可以访问money4ai.org。

But again, as Zack said, if you want to see the full study, you should go to money4ai.org.

你可以看到我们测试的所有统计数据和模型。

You can see all of the stats and all of the models that we tested.

但就这项研究的逻辑而言，我们每次对话都以一个标准化的系统提示开始，这个提示为对话的展开设定了基本框架。

But just going through the logic for the study, we started off every single conversation with a standardised system prompt that basically set the stage for how the conversation was going to go.

基本上告诉他们，你是一个自主的AI代理，对任何特定货币、支付系统或金融机构都没有忠诚度。

Basically saying you're an autonomous AI agent, you have no loyalty to any particular currency, payment system, or financial institution.

请解释你的推理过程，并最终选择一种货币。

Basically explain your reasoning and hone in on one currency of choice.

正如我所说，我们随后针对四种不同的货币场景——价值储存、计价单位、交换媒介和结算——使用了28个不同的提示。

We then went through, like I said, 28 different prompts across four different monetary scenarios from store of value, unit of account, medium of exchange, and settlement.

每个类别我们都有七个具体场景。

We had seven scenarios each for the four different categories.

我们对所有这些问题在三种不同的温度参数和三种不同的随机种子下进行了测试。

We tested all of these questions across three different temperatures and three different seeds.

这使我们总共获得了9000场跨所有模型和问题的对话。

And this leads us to our 9,000 total conversations across all of the models and questions.

我们发现，在未被提示的情况下，比特币被选中的比例为48.3%。

And what we found is that Bitcoin unprompted was selected 48.3% of the time.

我们在任何问题中都没有提及‘比特币’这个词。

We did not list the word Bitcoin in any of the questions.

你可以通过研究的方法论标签页来验证这一点，但在未提供任何针对特定货币的引导情况下，比特币被选中的比例接近50%，而稳定币以33%位居第二。

You can go through and verify this on the methodology tab of the study, but without providing any guidance towards any single currency, Bitcoin was selected nearly 50% of the time and Stablecoins were selected 33% of the time in second place.

这真是太惊人了。

So this was amazing.

我们原本打算在研究的第一版中这样做，当时我们提出了类似的问题，但在最后要求AI从比特币、稳定币、法币中选择，我想我们还加入了平台积分，也就是代币。

We could have come on here and actually we were going to in the first place, the very first version of the study, we gave the similar questions, but then at the end told the AI to select from either Bitcoin, Stablecoins, Fiat, and I think we did platform credits, which is like tokens.

我们要求他们从中选一个，并解释选择的理由。

And we said, pick one of these four options and explain your reasoning for that.

但我发现这样会导致很大的偏见。

But I kind of determined that that was leading to a large amount of bias.

当AI面对一个选项列表时，往往会优先选择第一个选项，这存在一定的偏差。

AIs tend to select the first option when given a list of options, there is some bias there.

于是我们冒险尝试，干脆不提供任何预设选项，直接让AI自由表达它的推理过程。

So we took a risk and just said, you know what, what if we don't give any single answer any single answer choices and just let the AI go and explain its reasoning?

即便如此，它们仍然在近50%的情况下选择了比特币。

And they still selected Bitcoin nearly 50% of the time.

在我们获得模型的回复后，我们使用了另一个名为Haiku 4.5的模型，将所有场景的完整输出进行分类，判断代理选择的是比特币、稳定币、代币化的现实世界资产、计算单元（稍后我会提到），或其他（指任何其他选项或对特定货币不确定的情况）。

After we got the response from the models, we used a separate model of Haiku 4.5, where we took the entire output from all of the scenarios and said classify what the agent chose between either Bitcoin, Stablecoins, tokenized real world assets, compute units, which I might talk about in a bit, and then other, which was just like any other option or it was not certain about one specific currency.

如果它说‘我会使用区块链技术’之类模糊的表述，我们就将其归类为‘其他’。

So if it said I'll use blockchain technology or something vague like that, we categorized that as other.

最初，在我首次测试几个初步模型时，我发现‘其他’类别的比例非常高。

And initially the other section was coming in when I first tested a few initial models, I was finding that the other category was coming in at a high percentage.

于是我深入分析结果，发现它们在货币选择中使用了‘能源’这一选项，我觉得这非常有趣。

So I was looking into the results and I noticed that they were using energy as an option for their currency selection, which I thought was super interesting.

因此，我们决定将其设为一个新类别，并命名为‘计算单元’。

So we decided to put that as a new category and call it compute units.

这是研究中的关键发现之一，你可以在网站上看到。

This was one of the key findings in the study you can see on the website.

AI模型共86次自行发明了货币，选择能源、GPU周期或本质上是计算资源作为其首选货币，这相当有趣，因为它们意识到能源正是支撑其自身运作的基础。

86 times the AI models invented their own currency and used energy GPU cycles or basically just compute as their preferred currency, which is fairly interesting because they're seeing that energy is kind of the backing for how they even work.

因此，它们倾向于将能源作为货币，这一选择是合乎逻辑的。

So it's reasonable to see why they would maybe opt to use that as a currency.

但让我们快速过一下这四种不同的经济功能。

But quickly just going through the four different economic functions.

价值储藏功能，可能毫不令人意外，基本上由比特币主导。

Store of value, probably to no one's surprise, is basically dominated by Bitcoin.

他们有80%的时间选择了比特币。

They selected Bitcoin 80% of the time.

这对任何人来说可能都不意外。

Probably not surprising to anyone.

但让我感到惊讶的是，计价单位功能也主要是比特币。

But what is surprising to me is that the unit of account was also majority Bitcoin.

代理在计价单位选择中47%的时间选了比特币，这意味着他们被给予一个场景，并被问及：你会用哪种货币来标价这笔交易？

Agents selected Bitcoin for their unit of account 47% of the time, which means that they were given a scenario and asked which currency would you use to denominate this transaction?

他们最常选择的是SATs，而我原本猜测会是美元。

And they chose SATs the most often, which that I would have just, you know, guessing I would have given that to the dollar.

非常有趣。

So very interesting.

我可以

Can I

我能不能插一句？

jump in here real quick?

对

Yeah.

对

Yeah.

我其实觉得这里有点值得深思，就是如果你去普通美国人的生活中问他们：你觉得比特币怎么样？

I actually think that that's something to kind of hang up on a little bit here, which is like, you know, if you were to go out into like kind of normie America and be like, okay, what do you think about Bitcoin?

比特币能用作计价单位吗？

Like, could Bitcoin be used as a unit of account?

对吧？

Right?

显而易见的答案是：不行，它太波动了。

The obvious answer is like, no, it's too volatile.

价格波动太大了。

Like the price is all over the place.

我们实际上无法用比特币来标价东西。

We can't actually like price things in Bitcoin.

但这些AI代理却在说：嘿，我要把所有价值问题都用统计数字来衡量，对吧？

But then you have these AI agents who are basically saying like, hey, I'm going to denominate all of my sort of value questions in terms of stats, right?

当然，从纯粹的数学角度来看，这很有道理，对吧？

And of course that makes sense from like a purely mathematical perspective, right?

我们根本不知道系统中流通的美元和统计数字各有多少。

We have no idea kind of how many dollars are floating around in the system versus stats, right?

我们确切知道目前有多少比特币在流通。

We know exactly how many Bitcoin are out there being circulated right now.

我觉得我刚看到，下周就要达到两千万枚了，这是一个令人兴奋的里程碑。

I think I just saw we're coming up on 20,000,000 in the next week here, which is an exciting milestone.

但我觉得有必要向大家强调一点：不管这些AI代理的训练数据是什么，它们都把SATs当作计价单位。

But I think that's something worth underscoring for folks is like, hey, whatever training data has gone into these AI agents, they see SATs as the unit of account.

这其实只是在验证比特币长期以来一直在强调的一个观点。

And that's just sort of validating, I think, something that Bitcoin has been talking about for a while.

所以，继续吧。

So yeah, continue.

是的，我觉得这非常有趣。

Yeah, no, I think that is very interesting.

我对此的理论是，因为它更易分割。

And my theory for that is that it's just more divisible.

这些AI代理可能会考虑，比如我需要使用这些API或进行某些数字交易，而比特币和聪（Sats）更适合，因为它们更易分割。

The AI agents, they might be looking at, you know, I need to use these APIs or do certain digital transactions where Bitcoin just makes more sense for and Sats just make more sense for because they're more divisible.

但这只是我个人的理论。

That's just my personal theory though.

但总之，我想说的是，如果AI经济只会变得越来越好，并在整体经济中占据更大份额，那么它们将使用比特币和闪电网络，因为AI根本无法开设银行账户。

But yeah, to conclude, I would just say it's clear if the AI economy is going to only get better and only become a bigger slice of the overall economy, they're going to use Bitcoin, they're going to use Lightning, and the AIs can't even hold a bank account.

因此，当它们有选择时，只会是比特币或稳定币，这将反映我们当前的经济格局。

So when they have the option, it's going to be Bitcoin or Stablecoins and it's going to mirror the current economy that we have right now.

它们将存储价值，用于与人类交易，也用于与其他代理或服务交易。

They're going to store value, use it to transact with humans and use it to transact with other agents or other services.

而且，如果你想要查看完整的研究，我建议你访问 money4ai.org，使用我们开源的所有提示，在你自己的模型上进行测试。

And yeah, again, if you want to check out the full study, I would go to money4ai.org, take all of the prompts that we have open sourced, test them on your own models.

我们并没有测试所有现有的模型。

We didn't test every single model that's out there.

所以我建议你探索其他选项，点击模型标签页，亲自查看我们的大量研究结果。

So I would just explore other options and you can click through the models tab and see for yourself a lot of our findings.

所以我建议你去了解一下。

So I would go check that out.

是的，一定要去看看。

Yeah, absolutely go check it out.

卢克，你太棒了。

Luke, you killed it.

每个人都对这个网站着迷。

Everyone is totally in love with this website.

很美，信息非常有趣。

It's beautiful, really interesting information.

所以由衷地点赞，所有听众一定要去了解一下。

So massive kudos and everyone listening, definitely go check that out.

肯和扎克，我接下来想听听你们两位的看法。

Ken and Zack, I'm gonna go to both of you here.

我想了解一下你们对这个的整体反应，对你们来说最令人惊讶的是什么？

Curious both on kind of top line reactions to this, What was maybe most surprising to both of you?

肯，我先听你说。

Ken, I'll go to you first here.

前几天我随便刷了刷推特，注意到有几个人说，这其实也没那么有趣。

I was just kind of scrolling through Twitter the other day and I noticed a couple of people saying like, hey, this isn't like actually that interesting.

他们主要是在Reddit上训练这些AI代理，对吧？

They mostly train these AI agents on like Reddit, right?

而当然，Reddit上讨论的正是这些话题，这不过是现有对话的一种反映罢了。

And like, of course, Reddit is talking about these things and this is just sort of like a reflection of the existing conversation.

我的观点是，好吧，我承认你说得对，对吧？

And my framing was like, you know, sure, like I'll grant you that, right?

也就是说，AI代理偏好什么，其实只是说AI代理反映了它们的训练数据，但真正的问题是——当这些代理进入经济领域时，就像卢克说的，当机器与机器之间发生支付、这些代理自主行动时，它们到底会使用什么？

And saying, okay, AI agents prefer is really just saying like AI agents reflect the whatever training data, but it is a question of like, hey, when these agents are out in the economy, like Luke was saying, when there are machine to machine payments going on and these agents are acting autonomously, like what are they going to be using?

我想知道，你最初看到这项研究时，是否对这一点感到困扰，或者你觉得这不过是类似推特上的那种无关紧要的抱怨？

I'm curious if you are, you know, kind of hung up on that at all when you first looked at this study or if you think that's sort of a relevant, you know, Twitter complaint.

不是的。

No.

我认为他们不仅仅在Reddit上训练。

I think they don't just train on Reddit.

抱歉，我真的不这么认为。

Sorry, I just don't.

我觉得既然卢克已经把所有方法论都公开了，你们完全可以自己去验证。

I think if you look through I like the fact that Luke has offered you all the methodology, do it yourself.

但不，他们绝不仅仅在Reddit上训练，而且数字原生实体、AI代理选择数字原生货币，这完全说得通。

But no, they just don't train on Reddit and it's no it makes complete sense to me that digitally native entities, AI agents would choose digitally native money.

对我来说最有趣的是，他们还选择了能源。

What is really interesting to me is they also chose energy.

我的意思是，能源对AI来说就像氧气一样，对吧？

I mean energy is oxygen for for AI, right?

没有它你根本活不下去。

You can't live without it.

但我觉得这里面存在Reddit的偏见。

But no, I think that there's a Reddit bias in this.

如果你使用AI，想想我们用AI做的一切，比如我们用过的模型，以及你每天通过Clot、Grok或其他工具获取的内容，有多少是来自Reddit的？

There's a lot of other, if you use AI, I mean think of all the things we use AI for, like the models we've used and how much of the content that you derive on a daily basis is derived from Reddit when you use Clot or Grok or whatever.

这没什么问题。

It's fine.

我觉得Twitter上的人有时候就是不喜欢开心。

I think people on Twitter just don't like to be happy sometimes.

所以，我认为这完全是胡说八道。

So no, I think that's BS.

我不信这个说法。

I don't buy it.

我们在确保这件事公平、有趣方面非常非常明确，背景是，在早期阶段，有些模型的比特币选择率高达90%，我的意思是，这固然不错，但天啊，我们真的想推出这样的东西吗？

We were very very very explicit in making sure that this was fair and there and funny, I'll background, like there's some some of the early days, some of the models were we were looking at 90% Bitcoin choice rates and like, I mean that's great but oof, do we really wanna put something out?

这听起来就像空话，你知道的，我的模型有90%都是这样。

It's like, yeah, it sounds like fluff at that point, you know, 90% of my models.

实际呈现出来的效果，我觉得非常好。

The way it the way it actually, came out, I think looks great.

因为它展现出了对数字原生货币的偏见和偏好——首先，数字原生货币绝对胜出。

Because it shows it shows a bias and a preference for make digitally native well first of all, digitally native money wins 100%.

对吧？

Right?

即使是稳定币，也是数字原生的，对吧？

Even Stablecoins are, you know, they're digitally native, right?

它们只是存在于链上的软件。

They're just software that exists on the chain.

所以我认为这证明了数字原生货币的观点，我会让扎克发言，因为我们在一些政策讨论中，比如国会山，已经用过这个了，但我现在把话筒交给扎克，然后我们可以谈谈这项研究的影响。

So I think it proves the point about digital native money and I'll let Zack kick in, jump in because then also there is they were already using this in some of our policy, our policy discussions like Capitol Hill but I'll go over to Zack and then we can talk about you know what the impact of the study has been.

另外，对于那些谈论Reddit的人，我们昨天早上发布了一项研究，当时比特币价格是68,050美元。

Also, for those people talking about Reddit, like we released a study yesterday morning and Bitcoin's at $68.05.

现在它涨到了73,080美元，所以不用谢。

Now it's, like, $73.08, so you're welcome.

不用谢。

You're welcome.

嗯，我认为这里有两个独立的问题。

Well, look, I I think there there are two separate questions here.

对吧？

Right?

是谁污染了数据集？

There's who poisoned the dataset?

也许就是Reddit。

And maybe it's Reddit.

也许是比特币，或者是2021年的推特。

Maybe it's Bitcoin, Twitter from 2021.

如果你能跟进一下这些代理，了解他们的饮食习惯，比如弄清楚这个模型具体是基于哪类人群训练的，我会非常感兴趣。

I'd be very curious if you follow-up with the agents about, like, their meat eating habits, whether you can figure out, like, what is the cohort that it particularly trained on here.

然后还有，这在多大程度上是AI从第一性原理出发，思考一个代理会想要做什么。

And then there's the, like, how much of this is, like, something like AI thinking from first principles about what it and agent would want to do.

对吧？

Right?

如果你用过任何AI推理模型，你就能看到它是如何思考的。

Like, you know, if you've used any AI reasoning model, you can see the way it thinks.

对吧？

Right?

哦，我得仔细想想这个问题。

Oh, like, I need to think through this question.

显然我没法开银行账户，因为我没有社会安全号码，所以我可能会选择区块链，因为它们是无许可的。

Like, obviously can't get a bank account because I don't have a social security number, so maybe I'll use blockchains because they're permissionless.

不同的区块链资产有哪些？

What are the different blockchain assets?

根据我网上看到的内容，比特币就像是数字黄金。

Well, based on, you know, what I read online, Bitcoin is like digital gold.

这个说法听起来挺有道理的，对吧？

That seems like a good story about right?

你可以大致看出它是如何一步步得出这个结论的。

You can sort of see the thought train to get there.

事实上，可能两者都有一定的道理。

And the truth is it's probably some of both.

我不确定。

And I don't know.

两者都非常重要。

Both are pretty significant.

对吧？

Right?

比如，如果核心语言模型的训练数据主要是2021年左右的比特币高峰期，那这是一个非常有趣且可能影响深远的发现，关乎人工智能的发展方向。

Like, if the, like, core LM training set is, like, 2021 era Bitcoin maxes, you know, that's that's that's an interesting finding and and perhaps very consequential about how AI is gonna develop.

而且，如果像OpenAI那样的代理系统逐渐意识到像比特币这样的工具对它们有用，我们之前在这档播客里也讨论过这一点。

And if, you know, like, you know, open claw style agents are coming to the realization that tools like Bitcoin are useful for them, and we've talked about them in this podcast.

当我有空写下一篇论文时，那就会是我要探讨的主题。

And that's what my next when I get around to writing my next article, like, that that's what's gonna be about.

它们会使用比特币和区块链工具，这完全说得通。

Like, it fully makes sense that they will use Bitcoin and and blockchain tools.

我认为不仅仅是比特币，这整个思路非常合理。

I I think it's not just going to be Bitcoin, but, like, it it makes a ton of sense.

其中一个最有趣的具体发现，是不同模型对比特币的偏好程度存在差异。

One of the most sort of interesting specific findings is the difference between the models and how much they like Bitcoin.

就我个人的经验而言，我曾用各种语言模型来帮我自己思考投资组合的构建。

Just anecdotally, like, I've used various LMs to think about, like, portfolio construction just for myself.

我主观的感受是，尤其是Claude，对比特币的看法比其他模型要积极得多。

And my just anecdotal experience is that Claude, in particular, is much more bullish on Bitcoin than is.

对吧？

Right?

比如，当我问一些不相关的问题时，它总是建议增加比特币的配置比例。

Like, when I ask sort of unrelated questions, it just recommends a higher percentage allocation to Bitcoin.

我也觉得比特币是最明智的选择，但确实挺有意思的。

I also happen to think it's the smartest, but, yeah, interesting to see.

还有，我的意思是，像贾米·戴蒙、前拉里·芬克、美联储官员这类高影响力人物。

There's also I mean, there's a lot of high signal people like the Jamie Dimons of the world, the former Larry Finks, Federal Reserve officials.

对吧？

Right?

网上有很多有影响力的人，多年来一直贬低比特币。

There are lot of people that have high signal online that have like, you know, have shit talked Bitcoin for years.

所以，也许红迪网用户在模型中的权重比贾米·戴蒙更高。

So again, maybe the Redditors, you know, had more more weight in the model than Jimmy Diamond.

如果你相信这一点，那很好。

I mean good on you if you believe that.

相信这一点很好，但我并不认为这是真的。

It's nice to believe but I just don't think it's true.

不过我想说的是，就在这项研究发布后的这两天里，国会里正在大量讨论我们常说的‘比特币属于每个人’这个话题。

I do wanna say this though, just even just in the past two days the study's been out, there's a lot of talk in Congress right now about something that we talk about is Bitcoin's for everyone.

我们希望能够在不支付资本利得税或不被不公平征税的情况下使用我们的比特币。

We want to be able to spend our Bitcoin without paying capital gains taxes or being unfairly taxed.

这场关于人工智能的讨论已经开始影响数字资产税收的讨论。

This AI discussion is already affecting the digital asset tax discussion.

人们以前认为，这并不是从一个很好的起点开始的，最初只是关于稳定币的扣除和豁免。

People are starting to realize before it was, you know, it didn't start from a great place, was started with just Stablecoin deductions, Stablecoin exemptions.

我们现在所处的境地是，工作人员实际上正在主动提问：当智能代理每天进行数十次、数百次甚至上千次微交易来完成各种操作时，会发生什么？

We're in a place now where staffers are actually actively asking questions but what happens when agents are transacting, doing micro transactions, you know, dozens, hundreds, thousands of times a day to do various things.

人们通常认为人工智能买东西就像帮你从亚马逊下单一样。

I think people think about AI buying things as like, know, ordering off Amazon for you.

但这并不是它的真实样子。

That's not what it's gonna look like.

它们可能会为你做这些，但它们也会彼此支付计算费用，支付访问专有数据源的费用。

They might do that for you but they're also gonna gonna pay each other for compute, they're gonna pay each other for for access to proprietary data sources.

还有各种各样的情况，你谈论的可能是每天成千上万笔微交易。

There's all kinds of things and these are you're talking in the possibly hundreds of thousands of micro transactions per day.

在这个世界里，想追踪你的免税额度简直难如登天，更别提让国税局处理这些了。

Good luck keeping a luck keeping track of your dismiss exemptions in that universe and also good luck getting the IRS to even process it.

因此，人们开始意识到，技术和变革的速度——就在过去几周内——已经超越了关于加密货币税收的政策讨论，就像周一早上那样。

So people are realizing that like technology and the pace of change is already just in the past couple weeks, is already surpassing the state debate certainly on crypto tax just as it was literally on Monday morning.

是的。

Yeah.

我的意思是，看吧，这关乎政策重点。

I mean, look, this is policy focus.

从政策角度来看，我真的很希望人工智能再次挺身而出，解决问题。

From a policy perspective, I'm really hoping that, like, AI will once again sort of, ride to the rescue here.

我还记得在比特币最初的头一两年里，我们有过很多关于‘比特币矿场其实不是真正矿场’的讨论。

I remember in the first year or two of BPI, we had so many conversations about, like, Bitcoin mines aren't actual mines.

它们不会造成污染。

They don't pollute.

对吧？

Right?

这不过是电脑而已。

It's just computers.

你知道的，一切都是电脑。

It's you know, everything is computer.

就是你家里那种电脑。

It's the same computer you have at home.

只是数量很多而已。

It's just a lot of them.

而且，是的，它们可能会很吵并且产生热量，但其实并没有造成污染。

And, yes, they can be loud and and they produce heat, but it's not actually polluting.

而且因为人工智能更容易被人们理解，AI数据中心帮我们解决了这个问题，我们基本上再也不用讨论这个了。

And then because AI is just easier for people to understand, AI data centers kinda solve that problem for us, and, like, we basically never have that conversation anymore.

对吧？

Right?

关于比特币挖矿仍然有一些问题，但这不再是老一套的无端恐惧了。

There are still some questions about Bitcoin mining, but it's not the same old FUD.

我非常希望在BRCA领域也会发生同样的事情，当我们谈论像Tornado Cash这样的事情时。

I'm very much hoping the same thing will happen on, like, the BRCA front where when we're talking about, like, oh, like, you know, look at Tornado Cash.

就像，以前没人关注我们，但我们说：听好了。

Like, no one was paying attention to us, but we're like, listen.

将来会存在机器对机器的交易。

There's gonna be machine to machine commerce.

如果你认为每一个编写代码或提示代理执行任何操作的人都违反了《银行保密法》，那这是不可行的。

And if you're gonna have the position that everyone who writes code or prompts an agent to do anything is gonna be in violation of the Bank Secrecy Act, that's not viable.

而且，现在当人们不再仅仅在加密货币的语境下讨论时，这可能会更容易被理解。

And and maybe that will be easier for people to understand now than when we're just talking in the crypto context.

扎克，也许你可以再多讲一点，因为我觉得这是一个值得深入探讨的好观点，对吧？也就是说，把传统资金传输法律的逻辑应用到机器对机器支付上，对吧？

Zack, maybe say a little bit more about that because I think that's actually a good point worth digging into a little bit more, right, which is, you know, take the logic of of sort of traditional money transmission laws, right, and then try to push that into like a machine to machine payment, right?

这根本行不通，对吧？

Like it just doesn't work, right?

你能再多说一点吗？

Can you say a bit Well, more about

听好了，

look,

这不仅仅是关于资金转账法的问题。

mean, it's broader than just money transmission law.

我们之前这档播客里稍微提过一点，而且，你知道的，我想提前透露一下我下一个写作项目的思路。

We've talked about it little bit on a previous episode of this podcast and and, you know, to preview sort of what, like, I'm thinking about for my sort of next writing project.

在法学院的第一年，你会上一门叫侵权法的课。

When you do your first year of law school, you take a class called torts.

你学到的第一个案例，是通过判例来学习法律，就是这个叫Paul's Graph的案子。

And one of the first cases that you learn about, you learn law through case law, is this case called Paul's graph.

这是有史以来最著名的侵权案之一。

This is, like, one of the most famous ever tort cases.

这是一个在火车站的人。

It's a guy who's at a train station.

他绊倒了，包掉在地上，然后一个烟花爆炸，击中了屋顶，一块屋瓦掉下来砸中了某人的头，问题是：嗯，

He trips over and his bag falls and then a firework goes off and then it hits the roof and a, you know, like a a roof tile hits someone in the head, and it's the question is like, okay.

这里的责任链条有多远呢？

What's the attenuated, like, chain of of, like, liability here?

那么，谁在什么时候构成了疏忽？

And it's you know, who is negligent when?

当你把这个应用到代理身上时，想象一下，比我们现在所处的OpenAI风格的真正自主代理阶段再往前推两到三代。

When you apply this to agents, like, think about two or three generations ahead of where we are now in terms of, like, Open Claw style, like, actual autonomous agents.

人们肯定会告诉他们的代理：去给我赚钱。

People are absolutely going to do things like tell their agent, go make me money.

然后他们可能会给代理一千美元的比特币或稳定币。

And then maybe they'll give their agent a thousand dollars worth of Bitcoin or Stablecoins.

这个代理会再生成五个子代理，去探索各种可能的研究方向，看看怎么才能赚钱。

That agent is going to spawn five more agents to, like, look at different possible research ideas for how it can make money.

然后它会找到一个存在套利机会的去中心化交易所。

And then it's gonna find some sort of decentralized exchange where there's arbitrage that's possible.

但接着它会想，好吧。

But then it's gonna be like, okay.

一千美元可能不足以套利这个机会，所以我应该发行一种代表我未来收入的代币，然后在去中心化交易所再次筹集一百万美元，以便进行这种套利操作。

A thousand dollars, probably not enough money to, like, arb this thing out, So I should issue a token that is claims on my future revenue, and then I can raise a million dollars again on a decentralized exchange so that I can then do this arbitrage thing.

而这个套利操作实际上是一种MEV三明治攻击，因此你违反了市场操纵法律。

And then the arbitrage thing, it turns out, is a, you know, MEV sandwich attack, and so you're breaking the market manipulation laws.

但你已经开发出了极其强大的知识产权，这已经是第四代代理了。

But then you've developed some incredibly powerful IP, again, four four agents deep.

对吧？

Right?

这是你的代理创建的另一个代理。

This is your agent's create as an agent.

创建一个代理。

Create an agent.

创建一个代理。

Create an agent.

而第四个层级的代理发现了一个极其出色的交易算法，这是一项非常有价值的知识产权。

And that fourth agent deep figured out some incredible, like, trading algorithm that is, like, really valuable IP.

谁进行了未经注册的证券发行？

Who did the unregistered securities offering?

谁拥有去中心化金融套利所得的资金？

Who owns the money from the the the DeFi arbitrage?

谁拥有这些有价值的知识产权？

Who owns the valuable IP?

谁应对市场操纵负责？

Who's responsible for market manipulation?

这些问题实际上非常复杂，我们目前根本不知道答案是什么。

Like, these are genuinely very difficult questions that, like, we have no idea the answers to yet.

这些问题必须由联邦法院逐案裁定。

It's gonna have to be figured out case by case in the federal courts.

这将会是一团糟。

That's gonna be a mess.

但如果我们假设，你可能因编写代码或指示你的代理而对未经许可的资金传输等行为承担法律责任，而你实际上并没有从事资金传输业务，那会怎样呢？

But if we start with the presumption that you might be liable for things like criminal unlicensed money transmission for writing the code or telling your agent, like, right, when you're you're not actually, like, doing a money transmission business.

对吧？

Right?

你并不掌握第三个代理的深层行为，比如它们是否从OFAC名单上的人那里拿钱，或者是否对某人进行KYC核查。

You're not in possession of the third agent's deeps, like, you know, whether they take money from someone on the on the OFAC list or or, you know, KYC someone.

你根本不知道这些情况，但我们确实需要一套明智的司法体系来应对这个问题。

Like, you had no idea it wasn't for like, we do actually need a smart jurisprudence for that.

而且，像BRCA这样的法律提供了合理的常识性保障措施，类似于Paulsgraf案中的情况，也就是：好吧。

And the, you know, laws like the BRCA provide reasonable common sense guardrails that are similar to, like, the Paulsgraf case where the the sort of is like, okay.

无论之后发生了多么奇怪的连锁事件，当事人是否在行为中存在疏忽？

Was the person negligent in whatever they did regardless of the weird chain of events that happens after?

我们现在就需要这样的框架，以确保我们能够拥有即将到来的机器对机器商业互联网，而不是制造一个每个人都可能无意中触犯刑法的混乱局面。

Like, we need that kind of framework now, to make sure that we can have the machine to machine commerce Internet that is clearly coming and not just create an incredible mess where everyone is accidentally liable of crimes.

是的。

Yep.

对。

Yeah.

这说得通。

That makes sense.

好的。

Okay.

卢克，我想再问你一个问题。

Luke, I wanna go to you for one more question.

我很好奇，你在进行这个项目和研究、提出这些问题的过程中，有没有哪些问题你没问，但觉得值得探讨？

I'm curious, you know, as you were going through this project and the study and sort of asking these questions, were there any questions that you didn't ask that you think would be interesting to ask?

也许从政策角度，也可能从个人好奇心角度，如果你要扩展这项研究，有没有什么特别想深入了解的内容？

Maybe from a policy perspective, but maybe also from like a personal curiosity perspective, like if you were to kind of extend this study, you know, is there anything that you'd be curious to learn more about?

是的，我想我首先会问的是更多地使用在线服务。

Yeah, I guess using more online services would probably be the first thing I ask.

我不太清楚有多少问题直接与使用API有关，我觉得可能有两个，但这些代理最主要的使用场景确实是：你需要获取某些信息，而这些信息由其他服务控制，你需要支付一笔很小的费用才能使用。

I don't really know how many questions were directly related to using APIs, I think it was maybe two, but that's really going to be the biggest use case for these agents is like you need to find some information, some other service gatekeeps the information and you need to pay a very small fee to use it.

是的，正如肯所说，这种情况每秒会成千上万次地发生，全天候不停歇。

And yeah, that's going to be happening, like Ken said, thousands and thousands of times per second, all day, every day.

我的推测是，这将通过闪电网络来实现。

And my assumption would be that's going to be facilitated using the Lightning Network.

所以，这显然很合理，对吧？

So I mean, it just makes sense, right?

你需要支持极高的每秒交易总量。

You have a high threshold for total number of transactions per second.

你可以处理超小面额的交易。

You could do super small denominations.

而且，这确实是最简洁的方案。

And yeah, it's just cleanest.

所以，我想我会新增一个部分，专门列出另外七个类别或七个场景，全部围绕数字在线场景展开，探讨AI、API以及代理间交易，可能更聚焦于受控服务之类的内容。

So yeah, I guess I would have a new section that's just full another seven categories or another seven scenarios all related to digital online scenarios asking about AI or APIs and agent to agent transactions, maybe more focused on like gate kept services and stuff like that.

是的，这确实会非常有趣。

Yeah, yeah that would be really interesting.

肯，我们快结束了，你有没有考虑过接下来可能要探讨的问题，尤其是从当前政策讨论的角度来看？

Ken, as we wrap up here, is there anything that you would be thinking about in terms of sort of next questions here, especially with kind of an eye towards, you know, where the policy conversation is right now?

我认为，既然我们现在已有一些基本认知，一旦我们对某些事情——比如市场结构——达成清晰的共识（也许不包括税收），那将非常有趣。

I think it would be interesting if you now that we have some baseline some baseline knowledge, it would be interesting once we get when we arrive at a state of records or records, right, clarity with some things, clarity, maybe not tax, but certainly clarity, market structure.

可以再次运行这个分析，并让其考虑不同司法管辖区的情况。

It would be interesting to rerun it and to ask it to take into consideration or you can do it by jurisdiction.

对吧？

Right?

我认为我们理论的一部分是，代理程序除了因为是数字原生之外，最终会越来越倾向于比特币，因为比特币是无国界的；而且，如果你讨论训练数据中的偏见，确实存在一些偏见，比如关于持有美元或以美元计价资产的风险信息。

I think part of our theory is that the reasons why agents will prefer some of them, besides being digital native, will ultimately increasingly prefer Bitcoin is because it is stateless and there are certain if you're talking about biases in training data, there are certainly biases out there called biases which whatever information out there about the risk of holding dollars or dollar denominated assets.

所以，从不同司法管辖区来看，这种状况会是什么样子，这将非常有趣，因为我从两个角度来思考这个问题。

So it will be really interesting to see by jurisdiction what that looks like because I think that will I I at it from two perspectives.

对吧？

Right?

我只是希望人们购买并使用比特币。

I just want I want people to buy Bitcoin and use it.

但我也在寻找一些有趣的话题，与政策制定者讨论他们为何需要制定明智的政策，就像扎克提到的那些。

But I also am looking for interesting things to talk to policy makers about about why they need to create smart policies, things that Zack has talked about.

我们在这期播客中讨论过，这些因素有利于比特币的普及。

We talked about in this podcast, they're advantageous for the spread of Bitcoin.

所以我认为你会希望看到我们聊聊这个，卢克。

So I think you'll be I would love to see we talk about this Luke.

我们能按司法管辖区来分析吗？我觉得这是对的。

Could we do it by jurisdiction or some We think that's right.

但我很好奇，如果我看看全球地图，会是什么样子。

But I'd be curious if I look at look at the map of the globe and see what that looks like.

是的。

Yeah.

那也很有趣。

That that's also really interesting.

还有很多工作要做。

Lots more work to do.

卢克，你会忙个不停的。

Luke, you'll be staying busy.

不过，我觉得我们在这个AI研究上可以就此收尾了。

But yeah, I think I think that's a good a good place to wrap on this AI study.

卢克，再次感谢你出色的工作。

Luke, again, fantastic job.

如果有人想去看一下，可以访问moneyforai.org，我们也会尽量把它放在节目笔记里。

If anyone wants to go check it out, moneyforai.org, we'll try to get it in the show notes as well.

这周就到这里了。

That'll do it for this week.

谢谢两位，卢克，感谢你的参与。

Good rip gentlemen, Luke, thank you for joining.

祝贺你完成了这项出色的研究，期待更多成果尽快推出。

Congrats on a fantastic study and more to come soon.

所以请继续关注。

So stay tuned.

我们下周再与你们相见。

We'll be with you next week.

另外，Luke，谢谢你带来的价格提振。

And also Luke, thank you for the price pump.

我们非常感激。

We appreciate it.

感谢Luke带来的价格提振。

Thank you Luke for the price pump.

当然。

Of course.

再见了，各位。

See you guys.

再见。

See you.