CD167：PIP - 顶点信任网络

这样当你查询时，总能获得最新鲜的数据。

And then when you you query, you always get the most the most fresh data possible.

有道理。

Makes sense.

首先，我在直播聊天里看到Stacks了。

So first of all, I see Stacks in the live chat.

他刚才在问那个搜索引擎的域名是什么，是npub.world。

He was asking what the domain was for that search engine that is npub.world.

我看到弗兰克在直播聊天中回复了他，但我想特别说明一下——如果他在音频里都没听懂，那么播客听众就更不可能理解了。

I see Frank answered him in the live chat, but I just wanted to put it out there because if he didn't understand it on audio, then no one that listens to the podcast feed would understand either.

而且我们节目的绝大多数听众都是通过播客音频收听的。

And the overwhelming majority of listeners on the show is the podcast feed audio only.

好的。

Okay.

顺着这个思路，像Nostril这样的项目正在使用本地客户端网络信任机制。

So to pull on that thought, so something like Nostril is using local client side web trust.

我推测这还会增加额外计算量，更耗电量，就像验证签名等操作一样，处处都是权衡取舍。

I also assume that is adding additional compute and would hurt your battery life more just like checking signatures and everything else is just trade offs all the way down.

那么对你们来说，开发者要如何集成呢？

But so with you guys, how is a developer integrating it?

我在官网上看到提到DVMs。

I see a mention of DVMs on the website.

从开发者角度看，这种集成流程是怎样的？

How does that look from the developer perspective in terms of integration?

再从用户和开发者双视角来看，对Vertex的信任机制是如何体现的？

And then how does it look from both the user and the developer perspective in terms of trust in Vertex?

是的。

Yes.

没错。

Yes.

从开发者视角看，我感到非常自豪，因为我认为体验非常友好——所有关于算法、数据下载、分析验证等复杂逻辑都被封装了，开发者只需简单地签署一个请求事件即可。

So from the point of view of the developer, I'm very proud because I think the experience is really good because all of this complexity on algorithm and downloading data and analyzing it and verifying and so on, all of that is abstracted away and what you do is simply sign an event that is a request.

这个请求是一个主机事件，它内部包含参数，比如我想使用这个算法，我想使用那个参数等等。

So the request is an an host event, and it contains inside the parameters, like, I want to use this algorithm, I want to use this other parameter and so on.

你将其发送到我们的中继站，然后获取响应。

You send it to our relay, and then you fetch the response.

就是这样。

And that's it.

DVM服务已经以某种方式构建好了，基本涵盖了最常见的用例，比如你想验证安瓿瓶的声誉，就像你在个人资料上那样，在我看来这就是使用它的场景。

The services are the DVM services are already structured in a way that kind of covers the most popular use cases, like you want to verify the reputation of an ampoule, like you are on the profile, so that is like the place, in my opinion, to use it.

然后你可能有一个数据库，想存储一堆信息供以后使用，这意味着你将使用排名档案。

Then you want maybe you have a database, you want to have a bunch of information for later use, and you means you you are going to use rank profiles.

假设你有一个高级用户，想给他推荐内容；或者一个普通用户，你想说‘哦，你也应该关注这些人’，可以使用推荐关注功能，它会通过算法生成这些推荐，然后进行搜索。

Then let's say you have a premium user, you want to give him recommendations or maybe a regular user, you you want to say, oh, you should also follow these other people, you can use recommend follows that is going to, yeah, use the algorithm to come up with this recommendation, and then search.

如果你想在客户端添加一个漂亮的搜索栏，直接加上就行。

If you wanna have a nice search bar on your client, you can just add it.

在后台，这个搜索功能会询问我们的中继站，然后显示结果。

And in the background, that search just ask our relay and then shows the the the result.

在信任方面，是的，这是一个中心化服务。

Now in terms of trust, it is a, yes, a centralized service.

所以目前它是完全可信的。

So it's fully trusted at the moment.

唯一的保证是响应都经过签名，因此不可能存在中间人试图冒充Vertex的情况。

It's well, the only guarantee is that responses are signed, so there cannot be a man in the middle that that tries to impersonate Vertex.

如果有人冒充反冒充者，那倒是挺滑稽的。

That would be kind of fun funny that someone impersonates the anti impersonator.

有道理。

Fair enough.

而且实际上，为了减少不信任，下一步是实施我想到的方案——客户端验证机制。

And and, actually, the next step in order to minimize distrust is to implement something I have in mind, which is a client side validation schemes.

意思是说，假设我们在讨论非常遥远的声誉问题。

Meaning you get, let's say we are talking about very far reputation.

所以你会得到奥德尔的前五位关注者。

So you get the top five followers of Odell.

你想知道是因为需要确认这个奥德尔是本人还是冒充者。

And you want to know because you want to know if this Odell is the real one or an impersonator.

现在你能在客户端做的是：获取这五位后，查找你的关注列表和他们的关注列表，就能验证他们是否都关注这个奥德尔。

Now what you can do client side, once you have these top five, is to find your follow list and then their follow list, and then you can verify that all of them follow this ODAL.

虽然你无法证明这五人是四万六千人中最重要的，但能证明这五人确实关注了他。

So you cannot prove that these are the top five out of 46,000 people, but you can prove that those five actually follow that.

因此这种情况下，客户端验证能防止最坏局面——比如Vertex撒谎声称所有知名人士都关注这个假奥德尔，实际上却没有。

So you prevent, in this case, client side verification prevents the worst case scenario where Vertex just lies and and said, yes, all of these reputable people follow this fake Odell when in fact they don't.

关于信任机制还需注意：由于基于DVMs构建，意味着其他竞争者可以加入，希望由此产生的竞争能形成制衡。

And and also for trust, it's important to note that because it's built on DVMs, it it means that other competitors can come in and then hopefully there will be some competition that that keeps everyone in check.

但目前来说，还只是

But at the moment, it's only

那么在那个情境下，你大概可以直接查询两个提供商并在本地对比结果之类的。

So you got presumably in that situation, you could just you can hit two providers and compare the results locally or something.

对。

Yeah.

例如，我会选择你更信任的供应商。

For example, I'll choose the provider you trust more.

是的，那是一种可能性。

Yeah, that's, that is a possibility.

但你可以用一方审计另一方，这会是个有趣的信任模型。

But you could use one to audit the other would be an interesting trust model.

我是说，现在你处于一个位置上，只要你愿意就可以审查某人。

I mean, now you stand in a situation where you could presumably censor someone if you wanted to.

对吧？

Right?

我不是在指责你这么做。

I'm not accusing you of that.

我不认为你会这么做。

I don't think you are.

是啊，

Yeah,

我可以。

I could.

我可以。

I could.

明白

Got

了。

it.

只是为了明确一下。

Just just to be clear.

是的，我可以。

Yes, I could.

这是一项可信的服务。

It's a trusted service.

就像，我会说类似于主要的缓存服务模型。

Just like, I would say similar to the primary caching service, model.

百分之百。

100%.

某个人，其他人可以运行它。

Someone someone some other can can run it.

但是，是的，理论上，Primal可以。

But, yeah, potentially, Primal could.

大概，

Presumably,

对Primal模型最大的信任点与此类似，实际上就是Primal可以选择不向你展示某些内容，而你无从知晓，因为所有内容都已签名。

like the biggest trust with the Primal model is similar to this, which is in practice, which is that Primal could choose not to show you something and you wouldn't know Primal was choosing not to show you something because everything's signed.

你可以验证中间没有人在篡改数据，数据未被修改，但你无法确认是否一开始就未曾见过这些数据。

You can verify that there isn't a man in the middle happening, that the data hasn't been modified, but you don't know if you haven't seen the data in the first place.

是的，这可能就是数据扣留，我想它是这么说的。

Yeah, that could be data withholding, I think it says.

是的，这也是中继节点普遍存在的问题。

Yeah, that is also the case with relays in general.

比如，你无法确定那些GDT事件是否针对他们不喜欢的人。

Like, you don't know if they are GDT events of people they don't like.

在这一点上采用相同的信任机制，就像宿舍里随便丢纸条而你根本不会知道是谁。

Same trust model in that regard to like, yeah, like the dormitory, just drop notes and you just wouldn't know.

但他们无法修改笔记，因为你可以核对签名。

But they can't modify notes because you can check the signatures.

对。

Yeah.

没错。

Exactly.

而且，因为我们每项回应都有签名，那个事件你可以存储下来。

And also, because every response we make is is signed, and that event, you can you can store it.

这样就能证明我们过去行为不当，那会彻底搞砸我们，就像

And so you can that that can be proved that we misbehaved in the past, and that would fuck us, like

那你就完了。

Then you'd over.

是啊。

Yeah.

声誉问题。

Reputation.

对。

Yeah.

一个事件

One event

再次利用你。

use you again.

确实如此。

Exactly.

是的。

Yeah.

是的。

Yeah.

普通API不会出现这种情况，因为它们没有签名，至少通常不会。

And this doesn't happen with the normal APIs because they are not signed, at least not normally.

所以要有证据。

So have proof.

你不会有证据的。

You wouldn't have proof.

正是这样。

Exactly.

你不会有证据，这将变成各执一词的局面。

You would not have proof, and it would be my word against your word.

那样我们就无法做出判断。但在这个案例中，由于我们使用了DVMs，如果我们行为不当，所有人都会发现。

And then we would not be able to, like, decide or but instead, in this case, because we use DVMs, then, yes, if we misbehave, then everyone would would see that.

所以你这里有开发定价。

So you have the dev pricing here.

是的。

Yeah.

试用可获10,000免费积分。

Get 10,000 free credits is the trial.

看起来最便宜的方案是每积分半美分。

And then it looks like the cheapest scale one is half a cent to credit.

它是什么样子的？

What is it like?

这实际成本究竟如何计算？

What does that actually relate in actual cost?

比如如果维特要在实践中将其加入紫水晶系统，成本会如何体现？

Like if if Vitter was to add this to Amethyst in practice, how does that look cost wise for him?

这完全取决于你的使用场景。

It really depends on what you use.

举例来说，我们以原始系统为例会比紫水晶更直观。

And, for example, if let's make an example of Primal instead of Amethyst, which easier.

原始系统有个数据库。

So Primal has a database.

对。

Right.

他们可能有数百万个MPUB和数不清的笔记。

And they probably have many millions of MPUBs and many, many millions of notes.

他们很可能浪费了大量资源存储垃圾笔记。

And they are wasting most likely a lot of some resources in storing notes that are just spam or yeah.

他们可以每天或每周连接顶点系统，为数据库每条记录请求排名。

And what they could do is that every day or every week, they connect to Vertex, and then for each amp up of their database, they ask the rank.

他们将这个排名存储起来以备后用，比如用于高级搜索。

They store this rank for later use, like to power search, for example.

如果排名太低，他们可以直接删除这些事件，因为他们认为人们对这个事件不感兴趣。

And then if the rank is too low, then they can simply delete these events because they say, oh, people are not interested in this event.

这是我的接力站。

This is my relay.

这是我的存储库。

This is my storage.

我更倾向于以更高效的方式使用它，所以我会删除这些东西。

I prefer to use it in a more efficient way, so I delete these things.

所以我提到的用户正在使用排名配置文件。

So the user I'm referring to is using rank profiles.

因此每1000个PAP密钥的成本不到1美分。

And so for each 1,000 PAP keys, it cost less than 1¢.

所以如果你有100万个密钥，那将花费你99美元。

And so if you have 1,000,000 keys, that would cost you $99.

用9美元，你就可以清理你的数据库。

With $9, you take your database and you clean it up.

也许你可以这么做。

Maybe you do it.

哦，这相当便宜。

Oh, it's pretty cheap.

是的。

Yeah.

可能我定价太低了。

Maybe I'm underselling.

是啊。

Yeah.

嗯，低价起步挺好的。

Well, it's good to start low.

以后随时可以涨价。

You can always increase the price later.

情况并非相反。

It's not the opposite.

你应该先高价起步，之后再降价。

You can you should start with a hike and then decrease it later.

我不这么认为。

I don't think so.

我是说，这取决于你卖什么，但如果是企业级产品的话。

I mean, depends what you're selling, but if you're selling, this is probably mostly B2B stuff.

这是个面向开发者的产品。

So it's like a developer focused thing.

所以你需要先扩大装机量获取反馈，之后再提价。

So you just want the install base and you want the feedback first and then you can increase price.

这就是为什么你们设置了免费档。

That's why I assume is why you have a free tier.

因为你想让开发者先实际试用，通过具体用例了解产品。

Is because you want a developer to actually play around with it and see how it works through his use case before.

你不希望支付环节成为摩擦点。

You don't want the payment to be a friction point.

归根结底，对于大多数好产品来说，你未来可以调低或调高价格，用户都会接受。

At the end of the day, for most good products, you can lower or increase the price in the future and users will be fine with it.

我觉得人们有点夸大其词，对原始价格考虑过度了。

I think people overstate that a little bit and overthink it on what their original price is.

我确实对我的定价考虑过度了。

I definitely over thought my pricing.

你说吧。

You go.

我们还没开始谈呢。

We didn't talk first.

不知道我想去哪里。

Don't know where I want to.

首先我很好奇，在Nostr上你有没有那种'上帝模式'的列表，能显示最受欢迎的账户？

So first of all, I'm curious, out of the do you have, like, do you have a god mode of that lists, like, the most popular accounts on Nostr?

你有Excel文档之类的吗？

Do you have, like, an Excel document?

你有电子表格能直接显示谁最受欢迎吗？

Do you have a spreadsheet that just shows who's the most popular?

不，你问这个是因为想知道自己是不是第一名吗？

No, you are asking because you want to know if you are number one?

不，我是说，我猜Dorsey肯定是第一名。

No, I mean, I assume Dorsey's number one.

我只是好奇。

I'm just curious.

好奇是谁

Curious who

前十名是谁。

the top 10 are.

我现在不知道，但我记得当时我在测试并查看排名，是全球排名，不是个人化的。

I I don't know now, but I remember when I was, like, testing and looking at the ranks, global global ranks, not personalized.

我想Damos是第一。

Damos, I think, is number one.

然后是Jeff Dorsey，因为基本上每个人都关注Damos，我想。

And then Jeff Dorsey Because everyone follows Damos, basically, I think.

如果你安装Damos，我想它会自动关注Damos账号。

If you install Damos, I think it automatically follows the Damos account.

有个小技巧。

There's a trick.

我开玩笑的。

I'm joking.

我想Damos、Jack，然后可能是你，也许你是第三名。

I think Damos, Jack, and then probably you, maybe you are number three.

有意思。

Interesting.

我是想打击一下你的自尊心。

I mean to to hurt your ego.

我会说足够谦逊。

I'll say humble enough.

我想

I think

足够谦逊。

Humble enough.

这对我来说挺有趣的。

It's just kind of interesting to me.

如果前50名中有相当多惊喜我也不会意外，尤其是Noster是非英语全球平台。

Wouldn't be surprised if there's a decent amount of surprises in the top 50, particularly since Noster is global non English language.

比如，我很好奇谁是最受欢迎的日语使用者。

Like, I wonder who is the most popular Japanese language person.

我完全没概念。

I would have no idea.

我甚至不知道那是谁。

I don't even know who that is.

我看不懂他们的语言。

I can't read their language.

是啊，我也不知道。

Yeah, I don't know.

老实说不知道，因为我还没有这个功能。

Honestly don't know because I don't have the this feature yet.

是啊。

Yeah.

GG前十名。

GG top 10.

我也这么认为。

I think so.

很可能。

Probably.

他是史上最佳。

He's the GOAT.

是啊。

Yeah.

五个名额。

For five.

MDK也在前列。

MDK also in the top.

林、杰克·马勒斯、杰夫·布斯。

Lin, Jack Mallers, Jeff Booth.

总是这些人，非常受欢迎。

Always the same, very popular.

我是说，这很有趣对吧？

I mean, that's the funny thing, right?

Primal平台有个可选的趋势算法，关于这个趋势算法和谁会上趋势榜有各种阴谋论。

So with Primal, we have this optional trending algorithm and there's all these conspiracies about the trending algorithm, about who's on the top of trending.

但实际上，在社交趋势领域这已经是世界上最透明的算法了。

And meanwhile, it is the most transparent algorithm in the world in terms of social trending stuff.

它是开源的。

It's open source.

你可以看到具体的权重分配。

You can see the exact weightings.

你可以清楚地看到它是如何运作的。

You can see exactly how it works or whatnot.

它最初为什么存在？

And why does it exist in the first place?

很多人反对算法，但它存在的原因其实是大多数人喜欢算法并觉得它们有用。

A lot of people are anti algorithm, but the reason it exists is because actually most people like algorithms and they find them useful.

算法需要透明且不具掠夺性，但这引出了一个有趣的问题，而且它是可选的。

They need to be transparent and not predatory, but it opens up an interesting question and it's optional.

这引出了一个有趣的问题：人们真正想要的算法可能和最受欢迎的笔记不同，而公告板可以选择展示这些算法。

And it opens up an interesting question where if maybe the algorithms people want and the cool part about announcers they can choose is maybe the algorithms people want aren't actually like, what are the most popular notes?

他们想看到底层的内容。

They want to see stuff underneath that.

这就是为什么原始趋势不是阴谋论。

And that's why with the primal trending, it's not a conspiracy.

只是因为如果更多人关注某人，其单个帖子就更可能进入人气排名。

It's just if more people follow someone, it's more likely that it's going to have the popularity rankings on individual posts.

是的，可以说缺乏多样性也是因为那个算法是全球通用的。

Yes, that also is a, let's say lack of variety is also because that algorithm is global.

所以你总是会发现那些全球顶尖的人物。

And so you find always the top global people.

可以说，这是针对用户定制的，或者换个词来说，是个性化的。

That could be made, let's say, relative to the user or, you know, another word personalized.

这也是更符合Nostra精神的，我认为。

That's also that that is more like aligned, I would say, to the Nostra spirit.

但实际操作中，这肯定需要更多的计算，成本也更高。

But practically speaking, it's definitely more, more computations, more costly also to do.

这也是为什么那个特定算法——个性化页面排名比全局页面排名成本更高，因为它对我们来说竞争更激烈。

That's also why the, that algorithm in particular, personalized page rank is more costly than global page rank, because it simply is more competition for us.

是的，实现起来要困难得多。

Yeah, it's much more difficult to implement.

我确实喜欢个性化内容，而且我认为Nostra在这方面有独特的优势。

I Yeah, I do like the personalized stuff and I think Noster's uniquely well positioned for it.

我看到Diana在YouTube上的评论，如果她也感到困惑，可能其他人也会有点搞不清楚。

I see Diana's comment in YouTube and maybe it was a little confusing for other people if she's confused as well.

关于Primal，她在问我Primal信任模型的事。

With Primal, she's asking me about the Primal Trust Model.

在Primal中，我们有一个缓存服务器，而大多数其他Nostra应用直接连接到中继。

With Primal, what we do is we have a caching server while most other Nostra apps connect directly to relays.

你可以把缓存服务器想象成我们运行的一个超级中继。

So you can think of a caching server kind of like a super relay that we run.

它会尝试收集我们看到的所有笔记，就是每一条笔记。

And it tries to collect all the notes that we see, just every note that we see.

我们这样做是出于性能考虑，同时也是为了隐私。

And we do that for performance reasons, but we also do that for privacy reasons.

这是因为当你在Nostr上阅读笔记或查看个人资料图片时，缓存服务器也在获取这些内容，它会抓取所有媒体和相关信息。

And that's because when you're reading notes on Nostr or even reading profile pictures, which the caching server is also taking, caching server takes all the media and everything.

当你在普通Nostrap上阅读时，实际上是在访问大量其他网络服务器。

When you're when you're reading that on a normal Nostrap, you're actually hitting a bunch of other web servers.

你会访问到托管媒体的任何网络服务器、托管笔记的任何服务器，以及Nostrap上的所有内容。

You're hitting like any web server that's hosting media, any web server that's hosting notes, anything on Nostrap.

而使用我们的服务时，你只需访问我们的缓存服务器。

And so with us, you're just hitting our caching server.

这样你只向缓存服务器运营商暴露了IP地址和使用行为。

So you're only exposing your IP address and your usage to the operator of the caching server.

这就引入了审查方面的信任问题——如果我们恶意操作，可以预先从你的订阅源中移除某些内容，让你根本看不到它们。

Now that adds a trust element in terms of censorship where we can if we were operating maliciously, we could remove things from your feed that you would never see in the first place.

我们无法修改你的订阅源内容。

We can't modify your feed.

我们只能直接移除内容，让你从一开始就看不到它们。

We can just remove things and never show them to you in the first place.

正如Pip之前所说，这与普通中继服务器的信任模型类似——普通中继同样可以移除内容。

And as Pip said earlier, that's a similar trust model as a regular relay where a regular relay could have removed could remove things.

使用普通Nostrap的优势在于你可以连接多个中继服务器，除非所有中继都丢弃了某条笔记，否则你不会错过。

Now the advantage you have with a normal nostrap or a typical nostrap is you might connect to multiple relays, they all would have to, be dropping notes for you to miss them.

我们的应对方案是：缓存服务器是开源的，你可以直接在用户界面切换到自己或他人运行的服务器。

The mitigation we have for that is the caching server is open source and in the UI itself, can go and you can switch it to one you run or one someone else runs.

就像Pip在他的项目中所说，最理想的情况是有多个实体运行缓存服务器，你可以轻松在不同服务器间切换。

As Pip said with his project, the ideal situation is we're hoping multiple entities are running caching servers and you can just easily flip between them.

在他们离开诺斯特之前，叛变行为时有发生。

Mutiny used to before they left Noster.

不过话说回来，关于原始信任模型这部分确实有点啰嗦。

But yeah, anyway, that was long winded on the Primal Trust model.

我看到有朋友建议原始系统应该整合顶点平台。

I see friends saying Primal should integrate Vertex.

我觉得这个提议挺有意思的。

I mean, that's interesting to me.

我是说，我们其实也可以自主研发一些东西。

I mean, I think we're also like we could do stuff in house.

我想确认下，顶点技术栈本身是开源的吗？

I guess is the Vertex stack itself is open source?

对。

Yeah.

当然是的。

Yeah, of course.

全都是开源的。

All open source.

其实我可能该早点提到，我是开源项目的受资助者。

In fact, I am a maybe I should have mentioned earlier, but I am an open source grantee.

噢，太棒了。

Oh, awesome.

这个我好像知道。

I think I knew that.

是的。

Yeah.

没错。

Yeah.

全是开源的。

All open source.

对，基本上你随时可以拿去自己运行。

And yeah, you can basically take it and run it yourself if you want.

我得说它没那么即插即用。

It's not as plug and play, I would say.

不仅因为你需要提供托管和运行的基础设施，还因为它需要你临时决定某些参数配置。

Not not only because you have to provide hosting and the infrastructure to run it, but because it it requires some, let's say, ad hoc, you choose of certain parameters.

初始化过程更像是一门艺术而非科学。

And it's more like an art than a science to, like, do the initialization.

因为初始化时，你要从某个节点开始爬取网络——比如FirtJaf和Pub，毕竟如果你在用NoSales，多少会隐式信任FirtJaf。

Because when you are initializing, right, and you you will crawl the network from a certain point, which was, like, FirtJaf and Pub, simply because it is you know, if you use NoSales somehow, you trust FirtJaf a little bit, at least implicitly.

然后我们以此为起点，爬取这份关注列表，社交图谱就随着这些人扩展。

And then we we go from there, and we crawl this forward list, so the graph expands with these people.

当某人排名很高时，系统就会递归地开始晋升流程。

And then, recursively, it starts to when someone reach a very high rank, then it gets promoted.

晋升后，我们会抓取你的关注列表，把新人加进来。

And when you get promoted, you we we fetch your follow list, we add new people in.

这就是个递归的初始化过程，没错。

And this is a recursive recursive initialization that yeah.

这这这不仅仅是为了重做而黑白分明，但如果你愿意，当然可以这么做。

It's it's it's not black and gray just to redo it, but, yes, if you want, you can do it.

你当然可以运行自己的Vertex。

You can run your own Vertex, of course.

对。

Yeah.

我是说，我们早期在Primal上就做过类似开放DVM订阅功能的选择。

I mean, we also we did like the early stages of just an open DVM feeds opt in on Primal.

你可以使用Primal应用，订阅由Vertex或其他算法驱动的热门DVM内容，或者个性化推荐。

You could use the Primal app and you can subscribe to a DVM feed that's trending powered by Vertex or something, or personalized trending or.

是啊，你看，

Yeah, look,

归根结底，我只是想把这个想法抛出来。

at the end of the day, I just want to put it out there.

Primal的目标是让你掌控自己的体验。

The goal, the goal with Primal is for you to control your experience.

我们不想规定你应该读什么不该读什么。

Like we don't wanna tell you what you should be reading and what you shouldn't be reading.

我们的目标是让用户完全掌控自己的使用体验。

Like the goal is for the user to have full control over their experience.

尽管存在各种阴谋论，但各方面仍有许多工作要做。

And there's a lot of work that needs to be done everywhere despite the conspiracy theories.

你现在有在广播任何DVM内容吗？

You broadcast any DVM feeds right now?

不，不，没有DVM数据源。

No, no, no DVM feeds.

没有。

No.

我们只有我刚才提到的那四项服务。

We we only only have the four services that I mentioned.

就是这样。

That's yeah.

可以说，Vertex的定位，正如你所说，更多是B2B。

The the let's say, positioning, I I see for Vertex is more, as you said, b to b.

是的。

Yeah.

因为现在确实有了微支付功能，理论上可以用来支付单次搜索。

Because it is true that now we have micropayments, which you can potentially use to pay for one one single search.

但问题在于，我认为这甚至会给资深比特币用户带来奇怪的体验。

The thing is that I think it creates a strange user experience even for seasoned Bitcoiners.

比如你下载一个新应用，得先给钱包充值才能搜索。

Like, you download a new app, you need to top up the wallet so you can search.

这听起来很反直觉，会让你思考那些本该自动完成的事情。

You know, it sounds counterintuitive, and it makes you think about things that should be, like, automatic.

你本不该纠结'这次搜索值不值90聪'这种问题。

You shouldn't think, is this search worth 90 shots?

就...我也说不清。

Like, I don't know.

我在使用应用程序时不应该考虑所有这些步骤。

I I should not think about all of those steps while I'm using the application.

没错。

So Right.

以Primal为例，在我看来，它的搜索排名算法采用的是粉丝数——Vertex也提供这种指标——但这基本上是最糟糕的选择，因为它根本不具备抗操纵性。

And for example, with Primal, I think it could be like, the way I see it now is that the search, the the ranking is using a followers count, which is, like, the which Vertex also provides, but it's like the the worst algorithm you can you can choose, basically, because it's not really civil resistant.

过去我曾展示过mpep.world的搜索与primal的搜索对比。

And in the past, I've I've, showed, like, how the search in mpep.world compares to the search in primal.

比如说，有

For example, there is

对。

yeah.

比如有个叫Nostalgia的乐队有超过10万粉丝，其中99%都是假的。

For example, there was nostalgia band that had more than 100,000 followers and about 99% of that is fake or something like that.

就像机器人刷的吧？

It's just like bots, right?

人们批量创建账号互相关注来破坏基础分析机制，对吧？

People spinning up end pubs and following to break the basic analysis, right?

正是如此。

Yeah, exactly.

这其实是个很难解决的问题——既要用关注关系判断信誉度，又得知道何时该忽略某些关注。

And in it's it's actually quite a hard problem to solve because you use follows to figure out who's reputable, but also you need to know when to discard some follows.

所以我之前提到过我们采用的网络递归发现机制：只从那些经过其他用户背书、已建立信誉的人那里获取事件。

That's why I mentioned earlier the the recursive discovery of the network that we do, and we only we only fetch events from people that have been promoted, that have been that have acquired reputation from other people.

我们拥有冒充检测功能，还有更好的信息流。

So we have impersonation detection, we have better feeds.

我们具备应用发现和验证功能，以及搜索功能。

We have app discovery and verifiability, and we have search.

对于像Vertex这样的平台，您认为还有哪些容易实现的用例？

Are there any other use cases that you think are low hanging fruit for something like Vertex?

我认为最灵活的是排名配置文件，你可以用它给多个MPUB排序，这在你的应用内部可以有多种用途。

I think the most flexible is rank profiles where you can rank a bunch of MPUBs because that can be used in so many ways inside your application.

举个很有趣的例子，假设Fountain发放统计奖励。

Like, I make a very interesting example, let's say, Fountain gives out stats reward.

我确信会有机器人试图钻空子。

I'm sure there are bots that try to game it.

但你可以设置规则，比如只给通过Nostra登录的用户发放这类奖励。

But what you could do is say, okay, I only give such rewards to people that that do Nostra login.

这同时也是激励人们使用Nostra的方式，比如让他们能在Fonten上获得奖励。

So also it's a way to incentivize people to use Nostra so they can get the rewards on Fonten, for example.

在他们登录后，只有排名超过特定阈值才能获得奖励。

And then after they log in, they can only get the rewards if their rank is higher than a certain threshold.

这个阈值要足够低，避免过多误判，既不会误伤真实用户，又能过滤掉近99%专门刷奖励的mPub——这些mPub通常新建账号，听几期播客就重复操作来获取统计。

A threshold low enough that it it doesn't do many false positives, so it doesn't discriminate real users, but it removes almost 99% of the mPubs that go to this game, basically, would spin a new mPub, listen to a few podcasts, and then do it again to get a few stats.

这也是个有趣的可能性。

This is also an interesting possibility.

提供些背景信息：Nostrad Band显示有4200万到4300万个mPub。

Just to give some context, Nostrad Band shows 42, 43,000,000 mPubs.

这就是他们中继站里的东西。

This is what they have in their relay.

他们看到了所有这些MPUBs。

They they see all these MPUBs.

在Vertex平台上，我们仅有35万个信誉良好的MPUBs，占比不到1%。

In in Vertex, we have 350,000 reputable MPUBs only, so it's less than 1%.

根据我们的分析，目前Ostrata上99%的MPUBs都是垃圾信息发送者。

So 99 of the MPUBs on Ostrata at the moment, according to our analysis are spammers.

我认为随着真实资金和利润涌入Noster，这个比例未来还会上升，垃圾信息制造者会抓住这个机会。

And I think this ratio is going to increase in the future as real money and profits come into Noster, then the opportunity and to spam is going to is going to be used, basically.

就像很多人发的垃圾内容，大部分会是AI生成的劣质信息。

Like, a lot of people spam most of the content would be AI slop.

因此，如何辨别真伪、决定存储内容、筛选用户展示内容将变得极其重要。

And so figure out figuring out what's real and what to store and what not to store and what to show to users going to be really important.

这些数据有公开渠道吗？

Does, is that data public anywhere?

那个...那个数量...

The what the the number of The

信誉用户与全部公钥持有者的比例？

reputable the reputable users versus total pubkey scene?

大多数平台都有统计页面，你可以在那里查看。

Well, the most of them has a stats page and there you can see.

但那是闭源的。

But there's a closed source.

你有自己的吗？

Do you have your own?

没有。

No.

从我自己的数据来看，就像每次响应中都会记录的那样，这里标注了我们数据库中笔记的数量。

From my own, like in every in every response, it is written this this number of how many it's called the notes, how many notes we have in the database.

这个信息很有用，如果你想要创建——我们把它添加到每个响应中，因为你可以利用这些数据来设定一个阈值。

And that is useful if you want to create we we add it to every response because you can use that data for coming up with a threshold.

明白了。

Got it.

基本上，这个阈值取决于图中有多少人。

Like, this threshold depends on how many people there are in the graph, basically.

这就是我们包含它的原因。

That's why we included it.

我今天早上检查过，是350。

And I checked this morning and it's three fifty.

知道了。

Got it.

350，但是

Three fifty But

这不一定是指活跃用户。

that's that's not necessarily actives.

对吧？

Right?

不，确实如此。

No, exactly.

不。

No.

它只是有名气，比如，像Microsailor或Adam Beck这样的例子。

It's only reputable, like, for example, like, microsailor or Adam Beck.

我认为他们

I think they

Balaji也来了，发了一个帖子就走了，他也算数，对吧？

Balaji also came in, made one post and then left and he counts, right?

对，没错。

Yeah, exactly.

这些确实计入那350个，因为他们有很多关注者。

Those do count in that three fifty, because they are followed by many.

但我觉得活跃用户数大概是这个数字的一半。

But yeah, would say active, and this is all like an estimate, but I think active would be half of that number.

所以17万或15万Noster的真实用户数看起来是准确的。

So 170, 150,000 real users of Noster seems like seems accurate.

明白了。

Got it.

是的。

Yeah.

我觉得这个数字差不多。

I think that sounds about right.

我认为当我们公开讨论时，确实应该对真实数字或确切数字持稍微不那么保守的态度。

I do think there's something to be said about when we're talking publicly about it being a little bit less conservative about the real number or the exact number.

我的意思是，这始终是个估计值。

I mean, it's always an estimate.

但我相当确定，人们使用的那些统计页面——比如'蓝天'这类没人能独立核实的来源——其数字简直夸张得离谱。

But, like, I'm pretty sure blue sky, whatever stat page that people use that no one can actually independently verify is the numbers are, like, incredibly inflated.

至于X或Telegram这类平台到底有多少用户，那完全就是'信我准没错'的状态。

And then who the hell knows, like, how many users there are x or telegram or something like that's just a complete trust me, bro.

有时我觉得我们对自己高标准要求是好事，但可能标准定得过高了，因为人们总会试图进行简单比较。

And sometimes I think it's good that we hold ourselves to a higher standard, but maybe we hold ourselves a little bit too high of a standard because people then try and compare them apples to apples.

这根本不是同类比较。

And it's not apples to apples.

比如我看到像So Cali这样的情况：他虽没注销X账户，但在X上发帖后，在Noster笔记获得的互动（真实互动，包括回复和提问）反而比X上多得多。

Mean, I see there are situations where So Cali, for instance, hasn't deleted his X account and he posts to X and he gets way more engagement on his Noster notes, like real engagement, like people replying to him and asking questions and engaging with him than he does on X.

但X号称有6.5亿用户，而我们只有10万左右。

But X supposedly has six fifty million users and we have like a 100,000.

所以这些数字根本对不上。

And so the numbers just don't add up.

这就像...到底怎么回事？

It's just, you're like, what's going on there?

就是

It's just

大家要记住这一点。

something to keep in mind people.

弗兰，我看到你关于签名和缓存服务问题的评论。

Fran, I see your comment about signatures and the problem with caching service.

Primal缓存服务会保持签名完整。

The Primal caching service leaves signatures intact.

签名会被传递给终端用户。

Signatures are delivered to the end user.

我认为目前Primal应用实际上还没有验证签名功能。

The Primal apps right now, I believe are not actually verifying signatures yet.

将来会有的。

Will in the future.

我们只需要让应用性能更好些。

We just need the app to get more performant.

光是保持现有稳定性就已经很吃力了。

It's been a struggle to just get it stable as it is.

而且很多其他Nostrap应用也没有验证签名。

And a lot of other Nostrap apps also aren't verifying signatures.

我想再次称赞Nostrap的做法，他们基本上提供了可选功能。

I like how Nostrap does it once again, to give them a shout out, where he there's basically an option.

设置中有个选项可以选择是否进行签名验证。

There's an option in settings whether or not you wanna do signature verification or not.

原因主要是性能问题，而且会显著影响电池续航。

And the reason is because primarily is a performance issue and also kills battery life.

不过没错，这就是我们的目标。

But yeah, that is the goal.

不过确实，Primal有个功能我特别自豪，就是你可以直接查看笔记。

But yeah, you can actually One of the features I'm actually really proud of with Primal is you can go to a note.

你只需点击那三个小点，就能复制原始数据——完整未处理的原始数据。

And if you just press the three little dots, you can press copy raw data, and that's the full raw data.

里面包含签名等所有内容。

You have with signature in there, everything.

你可以轻松保存、导出并验证这些数据。

You can just easily save that, export that, verify that.

这个小下拉按钮里必须有这个功能，这很重要。

It was an important thing to have in the little dropdown button.

继续说吧。

Go on.

你是要说什么吗？

You're going to say something?

对。

Yeah.

我想说的是，签名验证的成本确实是个实际问题。

I wanted to say that, like, the cost of signature verification definitely is is something real.

这会影响客户端性能并增加耗电，尤其是智能手机等低功耗设备。

It's something like that hurts the performance and the battery of clients, mostly, you know, on smartphone, low power devices.

所以我才特别看好之前提到的客户端验证方案，这几乎能实现两全其美。

That's also why I'm kind of bullish on this idea that I mentioned of client side verification, that you can have almost both the best of both worlds.

这是个折中方案：把验证工作委托给Vertex服务器，同时确保服务器至少不会恶意欺骗——比如把冒名顶替者说成本人。

It's a trade off straight in the middle, where you can delegate all this work to, in this case, the Vertex server, but check that at least the server is not not lying, for example, in way that would be bad, like saying that an impersonator is the real one.

这是最坏的情况。

This is the worst case scenario.

通过这种客户端验证方式，你就能避免最坏的情况发生。

And then with this sort of client side verification, you would be able to avoid the worst case scenario.

所以这确实是一种折中方案。

So it's, yeah, a trade off, I think, in the middle.

我认为Noster应用总体上应该或可以探索这个领域，因为很多应用直接选择把所有功能都放在客户端。

And I believe, in general, Noster apps should or could explore this area because many go straight into, okay, let's do everything client side.

然后你会发现应用几乎无法运行，因为加载任何内容都要耗费很长时间。

And then you find that your app is barely working because it takes forever to load anything.

嗯。

Yeah.

或者也许可以采用中间方案，比如Primal那样设置缓存服务器，虽然需要稍微牺牲一些信任度，但能获得更好的性能和用户体验。

Or or maybe so situation in the middle, yes, Primal, where there is a caching server there, where you obviously compromise a little bit on trust, but you get a lot more performance, a lot better user experience.

嗯。

Yeah.

Nostr最酷的地方在于你可以按自己想要的方式使用它。

I mean, the cool part about Nostr is that you can use it in any way you want.

希望能看到更多实验性的尝试。

And it'd be nice to see more experimentation.

虽然现在已经有很多实验了，但我倾向于支持更多不同折中模式的探索，尤其是在Nostr发展的现阶段。

There's a lot of experimentation, but in general, I lean to more experimentation and more different trade off models being experimented with than less, particularly at this point in Nostrad option.

我们应该测试所有可能性，不断尝试并从中迭代。

Like let's test out all the things, play with them, iterate from there.

我不认为存在一个最好的Nostrad应用。

I don't think there's a best Nostrad app.

我认为对某人最有用的Noster应用取决于他们的技术能力、使用场景和威胁模型。

I think the Noster app that's most useful for someone is depends on their technical competence, their use case and their threat model.

而这正是Noster的魅力所在。

And that's the beauty of Noster.

Noster的美妙之处在于用户最终拥有自主权，可以按自己的意愿使用它。

The beauty of Noster is that ultimately the user has agency and they can use it how they want to.

更多这样的内容。

More of that.

我本来还想说什么来着？

What else I was going say?

我想这对中继运营商来说也是通用的，你的服务在笔记管理方面可能非常有用。

I guess this also just in general for relay operators, your service could be very useful in terms of what notes I keep.

很多人讨论付费中继与免费中继，但或许折中方案是像Vertex这样经过垃圾信息过滤的中继。

A lot of people talk about like paid relays versus free relays, but maybe the middle ground is like a Vertex spam mitigated relay.

对吧？

Right?

是的。

Yeah.

没错。

Yeah.

实际上我写过一篇博客，在那里我解释了这个为开放中继设计的算法。

I actually wrote a blog post on where I, let's say, explained it is this algorithm I designed for for open relays.

所以那些目前不做任何垃圾邮件过滤的中继站，比如那些大型的、演示版的、原始版的、鼻腔乐队式的。

So relays that do not do any spam filtering at the moment, like the big ones, demos, primal, nostril band.

它们将会被垃圾邮件淹没。

They are going to be flooded by spam.

无论如何都会这样。

One way or another.

是的，完全正确。

Yeah, exactly.

情况肯定会恶化一百倍。

It's going to get 100 times worse for sure.

因为如果有人能从发送垃圾邮件中获利，人们就会不停地发。

Because if there is an incentive to spam, people will just spam.

所以你能做的是可以应用速率限制。

So what what you can do there is you can use the you can apply a rate limit.

这个速率限制很智能。

And this rate limit is smart.

它取决于用户是谁，事件的作者是谁。

It depends on who is the user, who is the author of the event.

因为作为中继运营商，你需要优化，你面临一个优化问题。

Like, because as a relay operator, you want to optimize, you have an optimization problem.

你希望多推送人们喜欢的帖子，少推送人们不喜欢的。

You want more of the notes people like and less of the notes people don't like.

你可以把事件作者的声誉作为一个启发式标准来使用。

You can use the as a heuristic, the reputation of the author of the event.

比如说，如果一群人关注奥德尔，很可能是因为奥德尔写的东西很有趣。

Like, if a bunch of people follow Odell, it's most likely because Odell writes interesting stuff.

所以你可以说，好吧，你有限速，但你有更大、更高的速率限制。

And so you can say, okay, you have rate limit, but you have larger, bigger rate limits.

如果你是无名小卒，那么你的速率限制会非常小，然后就无法再写了。

If you are mister no one, then you have very small rate limit, and then you cannot write anymore.

所以你每分钟可能只能发布一个事件。

So you can have, like, a one event per minute.

而在问题中是ODAL，你每分钟可以有100或1000个事件。

And in issue are ODAL and you have 100 or 1,000 events per minute.

你拥有基于信誉的速率限制。

You have reputation based rate limit.

博客的最后一部分是因为论文中的这些排名是由Vertex提供的。

And then the final piece of the the blog is because this this ranks in the in the paper are provided by Vertex.

那么如何避免让垃圾邮件制造者浪费你在Vertex上的一大笔钱呢？

So how can you avoid a spammer making you waste a bunch of money in Vertex, let's say?

比如，我创建10亿个密钥，所有这些密钥都写入你的中继站，你就会向Vertex发起大量调用，我会很开心，但你可能会不太高兴。

Like, I create 1,000,000,000 keys, all these 1,000,000,000 keys writes to your relay, so you make a bunch of call to Vertex, and then I would be happy, but you probably will not be super happy.

你可以做的是对新MPUB调用Vertex中继站的次数进行IP限速。

What you can do there is you can use IP rate limiting for how many times a new MPUB can call the Vertex Relay, basically.

所以每个事件的速率限制基于信誉，以及你能让我为排名支付多少新amp up，这基于IP。

So the rate limiting per event is based on the reputation and and how many new amp up can you make me pay for the rank, that is based on the IP.

通过这种方案，这种两步走的方法，我认为你可以做得很好。我做了一个成本分析，对于支付Vertex费用的防御方来说，比攻击方购买IP地址要便宜得多。

And this with this scheme, with this two step approach, I think you can get pretty good I I made a, let's say, a cost analysis, and it's way cheaper to use for for the defender that pays Vertex than it is for the attacker that buys IP addresses.

对。

Right.

这关乎成本，但在此情况下防守方更占优势。

It's a matter of cost, but it's the it's the defender is highly favored in this case.

首先，我认为这个设计非常巧妙。

What about, first of all, I think that's really clever.

这是个很好的权衡平衡点。

It's a nice trade off balance.

我特别喜欢基于信誉的速率限制机制，它巧妙解决了信任网络的核心痛点——那些尚未建立信誉的新用户该怎么办？

Particularly like reputation based rate limiting is really interesting because it solves that issue that people have with web of trust, the core issue, which is like, what about new users that don't have a reputation yet?

相当于说：新用户会在公共中继站受到温和的速率限制，直到积累足够信誉。

It's like, okay, well, they'll just be softly rate limited on public relay until they build up a reputation.

这样他们仍能发布内容。

So they can still get stuff out.

只是难以实施垃圾信息轰炸。

They just can't spam easily.

那些...抱歉我在看实时聊天忍不住笑出声

What about people that, I'm just laughing at the live chat.

他们说什么OnlyFans上线了之类

They're like OnlyFans did up.

共用IP导致的误判问题怎么解决？

What about the false positives of people using shared IPs?

我认为互联网通用最佳实践（尤其使用Noster时）是使用托管VPN服务，像Proton、Moldad或Obscura这种多用户共享IP的供应商。

I think a general good practice on the internet, and especially when you're using Noster, is to use a hosted VPN, a shared VPN that like a Proton or a Moldad or an Obscura that has a bunch of users that are all using the same IP address.

这实际上是个特性而非缺陷，因为你可以自建VPN，这样就不必信任任何人，但这也意味着你拥有固定IP地址。

And that's actually a feature, not a bug because yes, you can self host your own VPN and then you to, then you don't have to trust anyone, but that means you have a fixed IP address.

所以我其实更喜欢信誉良好的托管VPN服务，因为这样我就能和其他人共享IP地址。

So I actually like hosted VPNs from reputable providers because it means I have a shared IP with a bunch of other people.

这种情况你怎么看？

What about that situation?

因为刚才是不是有很多人从热门VPN里涌出来？

Because what did you just get a bunch of people coming out of the popular VPNs?

然后你们就会被封禁？

And then you'd end up block?

是类似MPUB加IP的模式吗？

Is it like MPUB plus IP?

是某种组合形式吗？

Is it like a combination?

不是的。

It's not.

两个步骤是这样进行的。

The two steps are this in this way.

新事件进来了。

A new event comes in.

如果我掌握作者等级，就会基于MPAB实施速率限制。

If I have the rank of the author, then I apply MPAB based rate limiting.

抱歉。

Sorry.

基于信誉的速率限制。

Reputation based rate limiting.

如果我没有作者的等级，可能是因为这是一个新的MPUB，或者我尚未请求该等级，那么我会将这个POP密钥放入队列，然后这个IP只能放入100个，或者说，一定数量的新POP密钥，在此之前我从未在这个队列中见过它们，直到它们被速率限制。

If I don't have the rank of the author, because maybe it's a new MPUB or maybe I haven't asked for this rank yet, then I put this POP key in a queue, and then this IP can only put 100, let's say, or a certain number of new POP keys that I haven't I've never seen in this queue before it gets rate limited.

它。

It.

在这种情况下，速率限制意味着你无法将新密钥放入此队列。

Rate limited in this case means that you cannot put new keys into this queue.

那么这个队列将如何使用？

Then how is this queue going to be used?

你取出一千个这样的POP密钥，一批一千个密钥，然后你一次性向Vertex请求这些等级。

You take one thousands of these pop keys, a batch of those one thousands keys, and then you ask Vertex for the ranks in one single call.

这样你就能节省一些费用，因为这是一个批量请求。

And so you you save some money because it's a batched request.

因此，如果有大量用户通过VPN接入，而你从未听说过其中任何一个，你只需对其中几个进行评级，然后基本上就会慢下来。

And so if a lot of users come in from VPN and you have never heard of any of those, you just rank a few of them and then and then it would slow down basically.

然后过一段时间，根据你如何配置这个限制，他们将能够基本上向中继写入。

And then after some time, depending on how you have configured this limit, they will be able to to to write basically to the relay.

所以，是的，这绝不是完美的解决方案，但对于免费的中继来说，没有完美的解决方案。

So, yeah, it's not a perfect solution by any means, but there are no perfect solution for, like, free free relays.

你可以使用工作量证明，但那样你的智能手机就得与ASIC竞争了。

It's you could use proof of work, but then you would have your smartphone trying to compete with an ASIC.

是的。

Yeah.

工作量证明的证明行不通。

Proof of proof of work doesn't work.

或者你

Or you

必须付费，但那样就不再是免费中继了。

have to pay, but then it's not a free relay anymore.

这会产生一种奇怪现象，比如新用户第一件事就得付费。

And it creates a strange, like you're a new user, you have to pay, like first thing.

是啊。

Yeah.

我们不能让新用户第一件事就是付费。

We can't have new users paying first thing.

如果用户还没开始使用就得用比特币支付，我们永远无法获得大量用户基础。

We'll never we'll never get we'll never get substantial user base if they have to pay in Bitcoin before they even get started.

在他们甚至还没拥有比特币之前。

Before they even get Bitcoin.

没错。

Exactly.

哦对，我喜欢这个。

Oh yeah, I like that.

我觉得这很聪明。

I think that's clever.

这是个聪明的权衡平衡。

It's a clever trade off balance.

你还能想到其他什么用例吗？

Any other use cases you have on top of your head?

我们已经讨论过很多了。

We've gone through a bunch already.

是的。

Yeah.

所以我们提到了搜索推荐、清除垃圾信息、保护免费中继。

So we said the search recommendations, removing spam, protecting free relays.

前端那个，比如说防止把钱给那些自动化流程的机器人。

The the front end one, the let's say, protecting from giving out money to bots that automate the process.

对。

Yep.

我没有更多想法了。

Not I I don't have more ideas.

我觉得我们这类事情做得...这些就是

I think we did this kind of Those are

很多了。

a lot.

你不该有压力。

You shouldn't feel pressure.

我只是想确认我们没有漏掉一些容易实现的目标。

I just wanted to make sure we didn't miss some low hanging fruit.

你对当前nostr生态系统的整体诊断或看法是什么？

What is your overall diagnosis or thoughts on the nostril ecosystem right now?

我知道很多人都感到精疲力竭。

I know a lot of people are feeling burnout.

最近Nostra上有很多看跌情绪。

There's a lot of bearish sentiment Nostra recently.

你觉得我们现在处于良好状态吗？

Do think you we're in a good place?

你认为我们需要在哪些方面改进？

Where do you think we need to improve?

我认为我们目前状态不错。

I think we are in a good place.

比如Nostra虽然没有增长，但也没有萎缩。

Like, the fact that Nostra is not growing is, but it's not shrinking also.

我觉得我们正处于平台期，这意味着用户留存率是稳定的。

It's it's I think it's we are in a plateau, and it means that, like, if you are in a plateau, it means that there is retention.

从个人体验和定性角度来看，我认为Nostra已经比X平台好很多了。

And a lot of, like, from personal experience or more qualitatively, I think Nostril is much better already than X.

问题始终在于网络效应，目前规模还太小。

The problem is always the network effect is so it is too small at the moment.

但我认为随着新用户涌入方式的变化，这种情况会改变。

But I think this will change exactly how new users are going to come in, like, where the new waves would come from.

我完全不知道。

I have no idea.

我认为社区方面有很大潜力。

I I think there is a lot of potential in, like, communities.

是啊。

Yeah.

而且我还没看到一个令人信服的解决方案。

And I haven't seen a very convincing solution.

对正在研究这个问题的人说声抱歉。

Sorry to anyone who is working on it.

但就我个人而言，你知道，我实在找不到一款能放心推荐给朋友、用来替代Telegram或特定主题社区的应用。

But just my, you know, my my my honest opinion, I I don't have an app that I can recommend to my friends that they can use as a replacement for, like, Telegram or or their community for a specific topic.

就是说，我找不到能自信推荐的产品。

Like, I don't have something that I can confidently recommend.

我认为这里存在巨大机遇。

I think there, there is a lot of opportunity.

同时这会非常有趣，因为我认为这能解决信任网络的最大难题之一——冷启动问题。

And also, it's it's going to be also quite interesting because that I think would solve one of the biggest problem of web of trust, which is like the cold start.

比如在不依赖全局算法的情况下（虽然总能用算法），你该如何在尚未关注任何人时查看内容并获得推荐？

Like, how do you, like, how do you how can you use without using global algorithms, which which can always use, but without that, how can you, you know, see content and get recommendation if you haven't even started following anyone?