Vitalik：區塊鏈只是工具，究竟要保護什麼？

0:00區塊鏈是工具，CROPS 才是目標

0:00好，謝謝，大家早安。今天我想談的，是我對區塊鏈思考方式的一個重要轉變，我覺得很多人也開始這樣想以太坊：它在世界上的位置、該怎麼用、以及不該怎麼用。標題是「區塊鏈是一種工具，CROPS 才是目標」。Okay, thank you, and good morning, everyone. So, so what I wanted to talk about today is what I think is an important transition in how I have been thinking about blockchains, and I think, really, how a lot of people have been recently, like, thinking about Ethereum, its place in the world, and how to use it, and also how not to use it. So. The title is "Blockchain is One Tool, and CROPS Is the Objective," right?

0:49現場有誰熟悉 CROPS？抗審查與抗俘獲、開源、隱私、安全。這些才是目標。區塊鏈是一種可以幫我們達成 CROPS 的技術。也有其他技術可以幫我們達成 CROPS，有時候用區塊鏈的方式，其實完全沒有幫任何人達成 CROPS。這是兩件不同的事。我覺得我們必須先從目標出發，不要太從某一種特定技術的視角看事情。大概五、六年前那個時代，我們犯過的錯誤之一，現場有人記得那個時代嗎？So, who here is familiar with CROPS? Censorship and capture resistance, open source, privacy, and safety. Those are the goals. Blockchains are a technology that can help us achieve CROPS. Blockchain. There are also other technologies that can help us achieve CROPS, and sometimes there are also ways to use blockchains that do not help anyone achieve CROPS at all. And so these are two different things, and I think it's important for us to really start with the goal in mind, right, and not kind of take too much of a perspective of focusing from the point of view of one particular technology, right? So I think one of the mistakes that we made in the era around five or six years ago, you know, who here remembers that era?

1:43有誰記得，把數位猴子賣到一千萬美元？那些極大、極華麗的研討會、Lamborghini。Justin Sun。那個時代的弱點之一，是我們把區塊鏈看得太像目的，不夠像手段。個人檔案，放到區塊鏈。身分，放到區塊鏈。社交圖譜，放到區塊鏈。NFT。NFT 好是因為它在區塊鏈上。如果一模一樣的東西是 Twitter 或 Google 當產品推出，這裡每個人都會討厭它。Who here remembers, you know, selling digital monkeys for ten million dollars? You know, all of the, like, extremely big and flashy conferences, the Lambos. Justin Sun. Now, one of the weaknesses of that era, right, is that I think we saw blockchains a little bit too much as an end and not enough as a means, right? So your personal profile, put it on the blockchain. Your identity, put it on the blockchain. Your social graph, put it on the blockchain. NFTs. NFTs are good because they're on the blockchain. If NFTs, if the exact same thing had been launched by Twitter or by Google as a product, everyone here would have hated them.

2:30DeFi 好是因為它在區塊鏈上，不管預言機是不是完全中心化、橋是不是後門多簽。我們為什麼這樣想？我覺得兩個原因。一，因為數字往上走。二，因為我們其實沒有別的工具。區塊鏈是那把唯一的鎚子，所以所有東西看起來都像釘子。尤其是今年，那種風格的加密貨幣基本上走到盡頭了。有一個原因值得誠實講：區塊鏈已經不是最亮的技術。DeFi, it's good because it's on the blockchain, regardless of whether or not the oracle is completely centralized, whether or not the bridge is a backdoor multisig, and so on, right? And why did we think this way? I think two reasons, right? I think one, because number go up, and two, because we didn't really have any other tools, right? Like, the blockchain was the one hammer, and so everything looked like a nail. So I think, especially this year, that style of crypto has basically hit the end of the road, right? And I think it's good to be honest about one of the reasons why, right? Basically because blockchains are not the shiniest technology.

3:19當紅技術是 AI，CROPS 需求更急

3:19你去問大學裡的人、VC、大公司，或任何地方的人：最亮的技術是什麼？他們會說是 AI。但同時，區塊鏈底下那些目標、CROPS，是人們需要得跟以前一樣急、甚至更急的東西。對安全金融的需求，比以前更強。未來一兩年，傳統金融大概會有大量被駭。什麼基礎設施能讓金融同時開放又安全？基本上就是過去 15 年一直在把這件事做到好的那套，也就是區塊鏈。專有 LLM 則是中心化的卡點。If you go to someone in a university, or in a VC, or in a big company, or anywhere else, and you ask them, like, what is the shiniest technology? They're going to say that it's AI. But at the same time, the underlying goals of blockchains, CROPS, is something that people need as strongly as they ever have. The need for secure finance, even stronger than before. Over the next couple of years, probably huge amounts of traditional finance are going to get hacked. What is the infrastructure that can make finance both open and secure at the same time? It basically is the infrastructure that has been perfecting exactly that for the last 15 years, which is blockchains. Proprietary LLMs as centralized choke points.

4:14如果你今天的工作流程依賴 OpenAI、依賴 Anthropic，那就是卡點。隨時可以被拿走。對你使用 AI 的監控。你用這些模型，另一端就有人或 bot 在看你怎麼用，而且很可能拿去訓練。我們不知道他們有沒有。我們也不知道下一個模型的訓練過程會不會意外記住你的密碼。還有用中繼資料去匿名化。所以在網路上保護隱私，正在非常快地變難。30 年前，我們對這件事很樂觀。If you have a workflow today that depends on OpenAI, that depends on Anthropic, that is a choke point. It can get taken away from you at any moment. Surveillance of your AI use. If you're using any of these models, then there is someone on the other end, or some bot on the other end, that is looking at how you use it, that is quite possibly training on it. We have no idea if they are or if they're not. We have no idea whether or not the training process of the next model accidentally memorizes your password, and so on. Metadata-based deanonymization. So, protecting your privacy on the internet is becoming harder and harder very quickly. So 30 years ago, this was something that we were very optimistic about.

5:03很多人記得那句口號：「在網路上，沒有人知道你是一隻狗。」問題是現在，在網路上，光一份文件就能把你去匿名化，就算你試圖保持匿名也一樣。幾個月前我發過那個挑戰：我寫了一份跟以太坊有關的匿名文件，公開之後兩週內就有人發現是哪一份。即使我用中文寫，再請 Qwen 3.5 翻譯。所以就算你很用力，去匿名化還是很容易。問題在中繼資料。I think many people remember the slogan, "On the Internet, nobody knows you're a dog." The problem is now, on the internet, you can get deanonymized from even one document, and you can even get deanonymized from a document if you try to keep it anonymous, right? This is a few months ago, I published that challenge where basically I had written an anonymous document related to Ethereum, and I published it, and within two weeks someone else discovered which one it was. Even though I wrote that document in Chinese and I asked Qwen 3.5 to translate it. So even if you try hard, deanonymizing is very easy. Now, the problem is metadata, right?

5:52你看去中心化即時通訊，像 Signal。Signal 能保護你資料的內容。它不保護誰跟誰通訊、你發訊頻率的那張圖譜。有了現代 AI，從那些東西能學到很多。還有對安全、可信任資訊來源的需求。過去 10 年，網路基本上變成全面的資訊戰場。AI 強化的審查。審查變得更強。還有全面的實體監控。今年已經出現很大的反彈。很多人開始非常擔心：攝影機數量變多很多，能即時解讀這些資料的演算法也強很多。So if you look, if you think of decentralized messengers, something like Signal. Signal, it can protect the content of your data. It does not protect the graph of who is communicating with whom, what is the frequency at which you're sending messages. And with modern AI, from all of those things, you can learn a lot. The need for secure and trustable information sources. The internet has, over the last 10 years, basically turned into a full-on info war zone. AI-enhanced censorship. Censorship is getting more powerful. Also, universal physical surveillance. This year there has been a large backlash. A lot of people are getting very worried about both the much larger number of cameras that's going up, and also the much more powerful algorithms that are able to interpret all of this data in real time.

6:49開源權重模型與密碼學

6:49而且不只是攝影機。是無人機、是 Wi-Fi 熱點、是各種能被變成攝影機的裝置。這些是全世界很多人極度擔心的題目。就算你問一個 20 歲、覺得區塊鏈是上一代技術、區塊鏈很蠢的人，他們至少也會擔心其中幾件。所以這些目標不只還活著，它們比 10 年前更重要，這件事值得記住。在今天主流技術公眾的看法裡，也值得承認：領先的自由科技甚至不是區塊鏈。And also, it's not just cameras. It's drones, it's Wi-Fi hotspots, it's all kinds of devices that can be turned into cameras. These are topics that many people around the world are extremely concerned about. Even if you ask someone who is 20 years old, who thinks that blockchains are the last generation's technology and blockchains are stupid, they will be concerned about at least some of these things. And so the goals are not just still alive; the goals are even more important than they were 10 years ago, and this is something that I think is really worth remembering. Now, in the mainstream technical public view today, I think it's also worth acknowledging, the leading freedom tech is not even blockchains.

7:38領先的自由科技是開源權重 LLM。但同時，區塊鏈會留下來。區塊鏈很強。我覺得很多人大大低估了區塊鏈。密碼學會留下來。五年前我們只是區塊鏈的人。今天我們是區塊鏈的人，也是零知識證明的人，也是全同態加密的人。也許再過幾年，我們也會是不可區分混淆的人。區塊鏈跟密碼學彼此變好。它們大幅擴張彼此的能力。一旦有了零知識證明跟其他可以疊上去的技術，你能用區塊鏈做的事會大得多。The leading freedom tech is open-weight LLMs, right? But at the same time, blockchains are here to stay. Blockchains are powerful. And I think many people are greatly underestimating blockchains. Cryptography is here to stay. Five years ago, we were just blockchain people. Today, we are blockchain people. We are also zero-knowledge proof people. We are also FHE people. And I think maybe in another few years, we will also be indistinguishability obfuscation people. Blockchains and cryptography make each other better. They massively expand each other's capabilities. And what you can do with a blockchain is much greater once you have zero-knowledge proofs and other technologies that you can add on top.

8:27所以以太坊重要。以太坊極度重要，而以太坊社群一直在核心位置開發、加速的密碼學，同樣重要。目標優先，技術是結論。來看幾個例子。不要從以太坊社群裡的人的視角想。從世界上在意那些目標的人的視角想。很多人有的一個目標是：用最大隱私去用前沿 LLM。LLM 供應商看得到你的查詢。And so, Ethereum is important. Ethereum is extremely important, and the cryptography that the Ethereum community has been central in developing and accelerating is just as important. So objectives first, technology is the conclusion. So let's go through a couple of examples, right? Don't think about this from the point of view of someone in the Ethereum community. Think about this from the point of view of someone in the world who cares about the goals. So one goal that many people have is they want to use frontier LLMs with maximum privacy, right? So the LLM provider sees your queries.

9:14他們甚至可能不願意提供 TEE 推論。所以你沒有機密性，但你能拿到的是匿名性。這裡有一種管線，能在你用前沿 LLM 時，有意義地保護盡可能多的隱私。第一步：現在本地模型已經夠強，可以用本地模型做編排。就算你用遠端模型做很聰明的事，也可以用本地模型去想查詢該長什麼樣。你不必每件事都直接用遠端模型。我筆電上跑著 Qwen3-8B。它其實非常強。現場有誰用過 Qwen3-8B？Qwen3-8B？They might not even be willing to offer TEE inference. So you don't have confidentiality, but what you can get is you can get anonymity. So here is one kind of pipeline that can meaningfully protect as much privacy as possible while you're using frontier LLMs. So step one: these days, local models are powerful enough that you can use local models for orchestration. Even if you're using remote models to do very smart things, you can actually use local models to figure out what the queries are. You don't actually need to use remote models directly for everything. I have Qwen3-8B running on my laptop. It's actually extremely powerful. Who here has used Qwen3-8B? Qwen3-8B?

10:05用本地模型保護前沿 LLM 隱私

10:05好，有幾個人。不，它其實非常聰明。第一步，你可以用本地權重模型做編排，讓它們改寫你的請求，拿掉所有私人資訊，甚至拿掉你的寫作風格。誠實的結論是：那不是你的寫作風格。那是大家都在用的同一種 AI 流水帳英文。模型看到的就是那個。第二步：匿名網路。匿名網路能提供網路層隱私。第三步是 ZK API。這是進行中的專案，他們其實很快就會發布。ZK API 把 ZK-SNARK 跟以以太坊為基礎的支付通道合在一起，讓你向供應商或代理付款，而每次查詢彼此不相連。Okay, a few people. No, it's like actually very smart. So step one, you can use local-weight models to orchestrate, to make them rewrite your request in a way that removes all of your private information, even in a way that removes your writing style. So the honest punchline: it's not your writing style. It's the same AI slop English that everyone else is using. That's what the model sees. Now, step two: anonymizing networks. So anonymizing networks can provide network-layer privacy. Step three is ZK API. This is an ongoing project, and they will actually be releasing very soon. And ZK API combines together ZK-SNARKs and Ethereum-based payment channels, and they allow you to make payments to a provider or to a proxy where each query is not linked to every other query.

11:04這基本上是把 Zcash、Tornado 或 Railgun 那種大多數匿名性質合起來，但不需要每次付款都做一筆區塊鏈交易。最後再送給供應商。這是一條解決問題的管線。我們可以看到，區塊鏈只在其中一個位置以解法的身分出現。來看第二個例子。想像你在意公共安全，也在意面對面的實體隱私。於是你說：我要一台偵測暴力的攝影機，只有偵測到暴力才錄影、才轉送影片。So this is basically combining most of the anonymity properties of something like Zcash or Tornado or Railgun, but without requiring a blockchain transaction every time you make a payment. And then finally, send it to the provider. So this is a pipeline that solves a problem. And what we can see is, well, blockchains appear in one place as a solution. Let's try example two. So let's imagine you care about public safety. You also care about in-person physical privacy. And so you say, well, I want a camera that detects violence, and it records and forwards video only if it detects violence.

11:52否則，它立刻、馬上刪掉看到的一切。我們希望這件事是人們可以驗證的。以今天能做到最好的方式，要怎麼做？首先，你要一個可被驗證的開源權重 LLM。你要它跑在可信硬體上。整台攝影機都要是可信硬體。然後開源權重 LLM 做偵測。它有分類器：這是需要被偵測並轉送的東西，還是只是一般人在做平常的事，那就立刻刪。如果你還想再加一層保護，可以再加一個功能。Otherwise, it immediately, instantly deletes everything that it sees. And we want this to be something that people can verify. So how would you actually build something like this to the best that's possible today? So first of all, you want an open-weight LLM that can be verified. And you want it running on trusted hardware. You want the whole camera to be trusted hardware. Then the open-weight LLM detects. It has a classifier: is this something that needs to be detected and forwarded, or if it's just normal people doing regular things, then instantly delete. And then if you want to add another layer of protection, then you can add one more feature.

12:38你可以說：每次影片被轉送，跑在上面、被可信硬體驗證過的軟體，會發布一筆鏈上交易。在拿到這筆鏈上交易已被收錄的證明之前，它不會繼續錄。這做了什麼？給你一個公開計數器。一個公開可驗證的計數器，你真的看得到這台攝影機實際轉送過幾次影片。於是你得到強得多的保證。你可以有可被驗證、不是全面全景監獄的攝影機。這不是從「我有區塊鏈，我想把區塊鏈塞進所有東西」這個視角能做出來的。You can say, well, any time the video is forwarded, the software that's running on it, which is verified by the trusted hardware, is going to publish an on-chain transaction. And it's not going to continue recording until it gets a proof that the on-chain transaction was included. So what does this do? It gives you a public counter. It gives you a public verifiable counter, where you can actually see how many times this particular camera has actually forwarded video. And so you get these much stronger guarantees, right? You can have, like, you can have cameras that are verifiably not a total panopticon. Now, this is not something that you can construct from the perspective: I have a blockchain. I want to put a blockchain into everything.

13:30區塊鏈只在需要的地方出現

13:30這是從這個視角：我在意 CROPS。我在意底下那些目標。我想試著解這些問題。結果在很多位置裡的其中一個，區塊鏈以解法的一部分出現。所以區塊鏈有時只是解法的一小塊。但不是零，而且是重要的一塊。區塊鏈對金融很有用。區塊鏈對做出可信承諾很有用。區塊鏈也有用來驗證某件事沒有發生。例如你可以驗證：某個檔案最新的雜湊是什麼。你可以用它做版本控制，驗證你拿到的軟體版本，真的是開發者簽過名的那一版。This is from the perspective: I care about CROPS. I care about the underlying goals. I want to try to solve these problems. And it turns out that in one place out of all of the many places, blockchains appear as part of the solution. So blockchains are sometimes a small part of the solution. But they're a non-zero part, and they are an important part. Blockchains are very useful for finance. Blockchains are useful for making credible commitments. Blockchains are also useful for verifying that something has not happened. So you can use them, for example, to verify: what is the latest hash of some particular file? You can use them for version control to verify that you're actually getting the same version of software that the developers have signed off on.

14:23你也可以把它們當計數器，驗證某件事發生過若干次、而不是更多次。所以如果你把區塊鏈看成工具，即使你的目標完全不含區塊鏈，它還是會意外地常常出現。密碼學讓區塊鏈更好，這是我今天稍後會談的：證明奇點。區塊鏈最大的問題之一是擴容。擴容怎麼解？用零知識證明。EIP-8288：遞迴 STARK 記憶體池。個別使用者發布帶證明的交易。那些證明再被聚合成其他證明，最後整個區塊只含一個證明，驗證所有人的活動。You can also use them as a counter to verify that something happened some number of times and not more times. So, if you see blockchains as a tool, they will come up surprisingly often, even if your goals do not include blockchains at all. Cryptography makes blockchains better, so this is something I will be talking about later today: the proof singularity. Basically, one of the biggest problems with blockchains is scalability. How do you solve scalability? With zero-knowledge proofs. So, EIP-8288: the recursive STARK mempool. And so individual users publish transactions with proofs. Those proofs get aggregated into other proofs, and then the entire block only contains one proof at the end that verifies everyone's activity.

15:13共識層簽章聚合。以太坊 L1 會有數萬個驗證者。那些驗證者簽章會被聚合，聚合用遞迴零知識證明來做，也會上鏈。資料可用性根，同一件事。這整塊都被 AI 很用力地加速，而形式化驗證、AI 輔助的形式化驗證，讓這次加速是安全的。所以以太坊路線圖，就我們未來三年想做的事來說，企圖心其實大幅增加了。過去大概十年，路線圖上想做的事大約到 2030。Consensus layer signature aggregation. The Ethereum L1 chain will have tens of thousands of validators. Those validator signatures will get aggregated, and the aggregation will be done with recursive zero-knowledge proofs that also get included on-chain. Data availability roots, the same thing. This is all being heavily accelerated by AI, and formal verification and AI-assisted formal verification is making all of this acceleration safe. Right. So the Ethereum roadmap, it has actually massively increased in ambition in terms of what we want to do over the next three years. The roadmap, I think, for about the last decade, we've had things that we wanted to do until about 2030.

16:03現在不一樣的是，清單大概變寬了五倍或十倍。我們怎麼做得到？因為 bot 在做很多寫碼。讓 bot 寫任何碼怎麼可能安全？因為我們有形式化驗證在驗證它。那你們能怎麼幫忙？台灣的人能怎麼幫忙？各地的以太坊社群能怎麼幫忙？我覺得重要的一件是，先回到根。CROPS：抗審查與抗俘獲、開源、隱私、安全。這是 2013 年之前的加密貨幣。有人記得 2017 年那篇文章嗎？The difference now is that the list of things has gotten about five or ten times wider. How are we able to do it? Well, because bots are doing a lot of the coding. How is it possibly safe to have bots do any of the coding? Because we have formal verification that is verifying it. So how can you help? How can people in Taiwan help? How can the Ethereum community everywhere help? So. I think one thing that's important is you want to start by going back to the roots. So, CROPS: censorship and capture resistance, open source, privacy, and security. This is crypto as it was before 2013. Does anyone remember, there was this article in 2017?

16:52回到 2013 年前的目標

16:52它寫：「每個人都可笑地變有錢，而你沒有。」我想你們可能記得 2025 年在 Twitter 上，有人拿同一篇文章開玩笑重貼，只是換成 AI 的人在變有錢。我們需要記住的是：長期真正要緊、超越這些誰今年變有錢、誰今年沒變有錢的週期的，是底下那些目標。那些目標從頭到尾都一樣。如果你問 2010 年還只有比特幣時待在比特幣裡的人，他們的目標清單會長得很像這份清單。It said, "Everyone is getting hilariously rich and you're not," right? And, you know, I think you might have remembered on Twitter in 2025, people making fun by basically, like, reposting that exact same article, except it's, like, the AI people getting rich. So the thing that we need to remember is the thing, what matters for the long term, going beyond all of these cycles of who is getting rich in one year, who is not getting rich in one year, is the underlying goals. The underlying goals are the same as they have been the whole time. If you ask someone who was in Bitcoin—Bitcoin was the only crypto in 2010—what their goals are, their list would look very similar to this list.

17:39所以去找一個你在意、又牽涉這些性質的問題。我們這個時代，是企業主義跟國家主義的科技封建、以及所有人對所有人的永久資訊戰的時代。這種問題很多。人們在意的問題很多。選一個。我也會說：有人跟我們共享底下那些目標、卻不用我們的鏈，比用我們的鏈去做完全不相容目標的人，更像盟友。所以找一個你在意的問題，弄清楚你需要哪些技術。如果區塊鏈不在其中，你可以不靠區塊鏈來做。如果區塊鏈是答案的一部分，那就絕對用。So find a problem that you care about that involves these properties. Our current age is the age of corporatist and statist techno-feudalism and permanent info-warfare by all against all. There are many problems that are like this. There are many problems that people care about. Pick one. I would also say someone who shares our underlying goals without using our chain is more of an ally than someone who uses our chain for goals that are completely incompatible. So find a problem you care about, figure out what technologies you need. If blockchains are not one of them, then you can build things without a blockchain. If blockchains are part of the answer, then absolutely.

18:27而以太坊在那裡等你，而且每過六個月都更強。現在是強制的以太坊路線圖更新。不知怎麼，以太坊路線圖更新總是會出現在我講的內容裡，所以我們講快一點。擴容：下一次硬分叉 Glamsterdam，會讓更高的 gas 上限變安全。它讓平行驗證成為可能。它包含修 gas 價格、解拒絕服務問題。EIP-8288，我稍後會談。我希望能放進 H-star，會讓隱私協議跟抗量子簽章變便宜。ZK-EVM 未來三年慢慢推出，會讓更高的 gas 上限也變安全。And Ethereum is there for you, and it's getting stronger with every passing six months. Now, mandatory Ethereum roadmap update. I think, you know, people, like, somehow Ethereum roadmap updates always appear in what I talk about, and so we'll make this quick. So, scaling: Glamsterdam, the next hard fork, will make higher gas limits safe. It enables parallel verification. It includes fixes to gas prices that solve denial-of-service issues. EIP-8288, I'll talk about it later. I'm hoping we can include it in H-star, will make privacy protocols and quantum-safe signatures cheap. ZK-EVMs being rolled out slowly over the next three years will make even higher gas limits safe.

19:14所以是大幅擴容計算，用證明來做，而不把鏈撐胖。隱私。FOCIL 會進 H-star，會給所有交易、包括隱私協議交易，強得多的交易納入保證。Frames，也就是 EIP-8141，看起來會進 H-star，會讓隱私交易不必經過中介，你就不必再走有時在線、有時離線、多一層依賴、還多收一大筆手續費的公開廣播者。EIP-8288 也會讓隱私同時更便宜、又能抗量子。So massively scaling computation and using proofs in order to do it without bloating the chain. Privacy. FOCIL, coming in H-star, will give much stronger transaction-inclusion guarantees to all transactions, including privacy protocol transactions. Frames, EIP-8141, which looks like it's going into H-star, will allow private transactions to be intermediary-free, and so you will not need to go through public broadcasters that, like, are sometimes online, sometimes offline, add extra dependencies, take a huge extra fee. EIP-8288 will also make privacy both cheaper and quantum-resistant at the same time.

19:59擴容、隱私、安全的路線圖

19:59於是我們會有完整的堆疊：用同一套 ZK-STARK 技術，把隱私跟安全同時帶到以太坊共識層、以及上面的應用，再加上存取層的配套技術。Kohaku 專案還在繼續。幾週前我用過。你現在基本上可以用 demo 版。你可以把它當介面，經由 Tornado、Privacy Pools 或 Railgun 送交易。能用。下一步是包含私密讀取，可以用 TEE 加 ORAM，或用 PIR。還有匿名網路，讓你可以私下送出，然後是安全。And so we will have a full stack where we're using the same ZK-STARK technology to bring privacy and security both to Ethereum's consensus layer and to the applications on top of it, and then also companion technologies on the access layer. So the Kohaku project is continuing. So a few weeks ago, I've used it. You can basically use the demo version right now. You can use it as an interface to send transactions through Tornado or Privacy Pools or Railgun. It works. The next step is including private reads, which can be done either with TEEs and ORAM or with PIR. Also anonymizing networks, so you can send privately, and then security.

20:48所以：把所有東西形式化驗證、讓跑節點更容易、讓輕客戶端更容易。每個以太坊使用者預設至少該有一個輕客戶端。把所有東西抗量子化。這很明顯。還有存取層的配套技術。去中心化 UI、鏈上 UI、社交恢復，那邊進度很多，做出好得多的恢復方式，不必依賴控制整個錢包的單一金鑰。多簽錢包、對 CROPS 友善的 AI 工具、本地 AI 工具：都在發生。你也可以用參與來幫忙。So formally verify all the things, making it easier to run nodes, making it easier to have light clients. Every single Ethereum user should have at least a light client by default. Quantum-resistify all the things. This is obvious. And companion technologies on the access layer. So decentralized UIs, on-chain UIs, social recovery, a lot of progress being made there, making much, much better forms of recovery that do not depend on having one single key that controls your entire wallet. Multisig wallets, CROPS-friendly AI tools, local AI tools: all is happening. Now, you can also help by participating.

21:35AI 對協議研究、協議開發做過的一件好處是：任何地方任何聰明的人，要跟上現有最前線、並很快開始貢獻，變容易得多。已經在跑的 AI 競賽有。Better.codes 是一個，snark.fast 是另一個，基本上是在改善以太坊會用的證明系統。我們大概很快還會有更多競賽。用 AI 輔助做極端前沿密碼學研究。這件事我覺得對任何地方的研究員、密碼學家、數學家都真的開放，可以來幫忙。One of the benefits of what AI has done to, like, protocol research and protocol development is it's made it much easier for any smart person anywhere to quickly get up to speed with the existing latest frontier and quickly start contributing. So there are already-running AI competitions. Better.codes is one, and snark.fast is another, basically for improving the proof systems that Ethereum will be using. And we'll probably have more competitions coming soon. AI-assisted research on extreme frontier cryptography. This is also something that is, I think, really open for any researcher, any cryptographer, mathematician anywhere to help at.

22:26這包括既可驗證、又帶全同態加密、還要極快的 AI。現在你可以做可驗證的 AI 推論，也就是放進 SNARK 裡，開銷不到 10 倍。我的夢想是全同態加密也能做到一樣，甚至 SNARK 疊在全同態加密上也能。Diamond iO 跟本地混合，也就是兩大家族的混淆演算法。不一樣的，你可以試著發明一種全新風格的超快 SNARK。你也可以用 Arklib 這種東西，真的用數學證明它安全。Arklib、EVMASM、LeanVM，很多進行中的專案你可以貢獻，或者你也可以去接以太坊世界裡我們還沒夠注意的任何角落。This includes AI that's both verifiable and with FHE, making that extremely fast. So now you can do AI inference that's verifiable, so inside a SNARK, with less than 10x overhead. And my dream is that we can do the same for FHE, and even for SNARKs on top of FHE. Diamond iO and local mixing, so both families of obfuscation algorithms. Different—you can try to invent a totally new style of ultra-fast SNARKs. And you can use things like Arklib to actually mathematically prove it's secure. Arklib, EVMASM, LeanVM—a lot of ongoing projects that you can contribute to, or you can take on any corner of the Ethereum world that we're not paying enough attention to yet.

23:16問答：永續合約是不是好事

23:16所以即使在核心研究上，讓人進來、讓人貢獻的表面積非常寬。我覺得以太坊的下一個時代、接下來這幾年，區塊鏈會跟最先進的密碼學完全合在一起，變得極度安全、變得可擴容，並在所有說得通的地方成為有用的工具，幫忙把更好的抗審查與抗俘獲、開放、隱私、安全帶到世界上、變成現實。我希望我們在座所有人，都能成為讓這個未來發生的一部分。謝謝。好，我們有問答。So even on the core research, there is a very wide surface area for people to come in and for people to contribute. So I think this next era of Ethereum, these next few years, as the blockchain fully merges with the most advanced cryptography, becomes extremely secure, becomes scalable, and becomes a useful tool in all of the places where it makes sense to help bring better censorship and capture resistance, openness, privacy, and security to the world into a reality. And so I hope that all of us here can be part of making this future happen. Thank you. All right, and we have some Q&A session.

24:07有問題請舉手，我們會把麥克風給你。我來戴上警長帽子。我們該用常識，想什麼是好問題。很遺憾我做不了西部牛仔口音。抱歉。用常識，問相關的、技術的，或真的關於改善 CROPS 這些事。既然我戴了警長帽，如果我覺得可能不是最好的問題，我會請工作人員跳過，好嗎？If you have any questions, please raise your hand, and then we'll give the mic to you. And I'll put on my sheriff hat. If there's—you know, we should use our common sense and then figure out what is a nice question. Unfortunately, I cannot do Western cowboy accent. Sorry, guys. But, you know, use common sense and just ask things that are relevant, technical, or, like, really about, like, improving CROPS or all these things. And since I have the sheriff hat, if I sense that it may not be the best one, then I'll ask the staff to skip, okay?

24:57好，第一個，那邊有一位。這邊我也看到幾位。謝謝。很高興在這裡見到你，歡迎再次來到台灣。我是 Ivy。我的問題是，以太坊上現在名列前茅的應用之一是 Lighter，我想問你對永續合約的整體看法。你覺得它對社會是淨正向嗎？你有預想過，推出十年後，這會是以太坊最好的例子之一嗎？謝謝。我覺得永續合約是工具。很強的工具。可以用來做很多事。確實有很多人只用它來賭博、把錢輸掉。也有人用它來對沖風險。So yeah, the first—there's one there. I see a couple here. Thank you. Very nice to see you here, and welcome to Taiwan again. My name is Ivy. And my question is, so one of the top applications on Ethereum right now is Lighter, and I want to ask, what's your general thoughts on perpetual futures? Do you see it as a net positive value for the society? And did you envision this to be one of the best examples of Ethereum, 10 years after launch? Thank you. Yeah, I mean, I think perpetual futures are a tool. They are a powerful tool. They can be used to do lots of things. There's definitely a lot of people that are just using them for gambling and losing their money. There's also people that are using them to hedge their risk.

25:53我覺得做出一套金融系統，讓人很容易在鏈上對沖各種風險，基本上更靠近這種完備市場，這個想法很強。我也感謝 Lighter 在用零知識證明、用更恰當的方式把它做得更可擴容上，做了很多研究跟實作。那很好。我幾個月前也發過一篇，建議我們該多用選擇權去實作穩定幣跟對沖這類東西。And I think the idea of creating a financial system that makes it, like, very easy for people to hedge all kinds of risks on-chain and, like, basically coming closer to having this kind of complete market is very powerful. So, I mean, I also appreciate that they've been, like, Lighter has been doing a lot of research on and, like, implementation on using zero-knowledge proofs and, like, making it more scalable in a more proper way. So that's been good. I mean, I know there's also, like, I published a few months ago this post where I suggested that we should be using options more to implement, you know, things like stablecoins and hedging.

26:40隱私友善監控如何落地

26:40我也知道已經有專案試著站在那個想法上繼續做。所以我覺得這些都可以很有價值。那邊有人舉手。謝謝。嘿，很高興你在這裡。你談到攝影機、硬體攝影機，基本上像在點名 Flock，我們要怎麼讓 Flock 基本上符合 CROPS，減少世界上的暴力，我們同意那是淨正向，但又不要全景監獄那種風格？目前沒有誘因讓任何人去實作這種東西。除了社會層有人在推隱私，而我們知道那行不通，還有嗎？And I know there's been projects that have been trying to build off on top of that idea too. So I think all of these things can be quite valuable. I see a hand right there. Thank you. Hey, nice to have you here. You talked about cameras and hardware cameras and basically, like, a shout-out to Flock, and how can we make Flock basically CROPS and reduce violence in the world, which we agree is a net positive, but without the panopticon style? There is no incentive currently for anybody to implement something like this. Is there, apart from the social layer, people pushing for privacy, which we know doesn't work?

27:26人們沒那麼興奮。我們有點像少數在推隱私的人。有哪些思路也許可以探索，好把這些帶到現實，不是技術面，技術面我們知道做得到，而是真正推動落地？這是好問題。我覺得答案有兩部分。一部分是，這些更隱私友善、更可驗證的做法必須成熟得多。它們必須是你可以直接部署、品質標準接近現有更主流解法的東西。必須到這種程度：對政治人物、或對管理大學這類環境的人來說，選擇就是簽另一家供應商，難度上沒有巨大差異。People are not so excited. We are kind of like a minority that push for privacy. What are some lines of thought that maybe we could explore in order to bring those to reality, not on the technical side, which we know is possible, but on the real push for implementation? Yeah, it's a good question. I think there's two parts to the answer, right? I think one part is that kind of these more privacy-friendly and more verifiable ways of doing things have to be much more mature. They have to be something that you can, like, just go and deploy with a similar standard of quality to some of the existing more mainstream solutions. Like, it has to be at the point where, for either a politician or, like, someone managing, maintaining an environment like a university or something else, like, the choice is literally just, you know, like, signing off on a different provider, and it's not like there isn't a large difference in, like, difficulty.

28:31我答案的另一部分是：如果你盯著底下那些目標，也不要低估我們有多少盟友。隱私是非常主流的擔憂，很多人想隱私的方式確實不一樣。確實有很多人對隱私當口號點頭，但你問他們匿名金融，他們會說那又壞又可怕，我覺得這是錯的。但最終，如果你回到這些目標，很多人擔心的是：誰能從攝影機看到他們，誰能重建他們所有移動、整段私人生活。I think the other part of my answer is, I would, like, if you focus on the underlying goals, I would also not underestimate how many allies we have, right? I think, like, privacy is a very mainstream concern, and a lot of people definitely don't think about privacy in the same way. There's definitely a lot of people, for example, who, you know, like, nod at privacy as a slogan, but at the same time they would be extremely willing to, you know, like, if you ask them about anonymous finance, they would say that it's bad and scary, which I think is a mistake. But ultimately, if you, like, think back to these goals, like, there is a lot of people who are concerned about, you know, like, who can see them through cameras, who is able to reconstruct all of their movements and their entire personal life.

29:27在機構層甚至更強。像 GDPR 這種法律，十多年來一直是歐洲科技環境很大的一塊。有非常多政府在意技術主權，不想讓所有公民的私人生活立刻被美國、或其他正在做這些技術的大國公司看到。所以即使在機構層，我覺得對這類事也有很多興趣。他們有時用的詞，聽起來比較像合規跟風險管理，而不是個人自由跟賦權。It's even stronger at the institutional level. Laws like GDPR have been a big part of, you know, like, the European tech environment for over a decade. There is a huge number of governments who are concerned about tech sovereignty and who do not want all of their citizens' private lives to be immediately visible by companies in the U.S. or in other large countries that are building these technologies. So even at the institutional level, I think there is also a lot of interest in, like, in these kinds of things, right? And they, you know, like, they sometimes use words that make it sound more like compliance and risk management than being about, like, individual freedom and empowerment.

30:13協議該不該更精簡

30:13但常常最終是同一批目標。所以我覺得，理解並做出這些更大的聯盟，也必須是答案的一部分。好，沒問題。嗨，演講很棒。我的問題是，十年之後……區塊鏈走過幾次迭代，我們其實沒有幾十種智慧合約原語。你覺得我們、或以太坊，該不該多學比特幣的 UTXO，把整個協議做得更精、更小？因為到頭來如果那是關於驗證、關於 ZK、關於……But often it's, like, ultimately the same goals, right? And so I think, like— Actually, understanding and creating these bigger coalitions also just has to be part of the answer. Yeah, yeah, that's fine. Hi, amazing talk. My question is, like, after 10 years... The blockchain goes through a few, like, iterations, and we don't really have dozens of primitives with smart contracts. Do you think we should, or Ethereum should, learn more, like the UTXO from Bitcoin, and then get the whole protocol more lean and minimized? Because at the end, if that's about verification, about ZK, about, like...

31:02證明，為什麼鏈上還需要圖靈完備的 VM？很多事可以在鏈下做，如果那件事仍然可以在鏈下實作、執行，區塊鏈本身為什麼還要自己處理整層 VM？原則上我同意。但覺得圖靈完備是壞事，我覺得有點走太遠。原因之一是，例如在以太坊協議裡，我們已經需要一種語言來描述遞迴證明聚合、共識層邏輯這類東西，而我們發現最務實的表達方式，就是有一個簡單、但仍然功能完整的 VM。Proof, why do we need a Turing-complete VM on-chain? Because a lot of things can be done off-chain, and if the thing can still be, like, implemented and executed off-chain, then why does the blockchain itself need to handle all the VM layer itself? Yeah, I mean, I think in principle I agree. Like, I don't think that saying, like, I think saying Turing completeness is bad is kind of going a little bit too far. And one of the reasons why is because, for example, like, we have already had a need within the Ethereum protocol to have a language within which to describe things like, for example, recursive proof aggregation, consensus layer logic, and we've discovered that, like, the most practical way to, like, express these concepts is by just having, like, a simple VM, but still, like, a fully functional VM.

31:54我其實希望長期來看，以太坊會有 VM，但是更簡單的 VM。RISC-V 目前是領先候選。我的希望是我們基本上處處用 RISC-V，既當使用者做出的證明的指令集，也當描述協議的指令集，甚至把 EVM 做成 IR，讓 EVM 不再是共識層的一部分。取而代之的是，再多一步從 EVM 編譯到 RISC-V，那就是一大塊程式碼可以搬家的地方。這也是我的目標之一。我真的希望接下來五年，以太坊協議可以開始縮小，變精簡得多。I actually hope that in the long term, like, Ethereum will have a VM, but it'll be a simpler VM. RISC-V is currently the leading contender. So my hope is that we basically use RISC-V everywhere, kind of both as the ISA for, like, proofs that users make, and for the ISA to describe the protocol, and even, like, basically make the EVM into an IR, and so the EVM stops being part of the consensus layer. Instead, like, you just have another compilation step from the EVM into RISC-V, and that's— like, basically where a big, a big bulk of the code can move to. So I, like, this is also one of my goals. Like, I actually really hope that the Ethereum protocol can kind of start shrinking and become much more minimized over the next five years.

32:45我可以問第二個問題嗎？好，第二個是關於穩定幣。我們現在知道 CROPS 跟 AI 重要，但穩定幣也許是過去十年加密貨幣最大的創新或採用，可是大多是 USDT 跟 USDC。那不是我們真正要的、不是無需信任的 ETH。現在看起來我們不再說 ETH 是錢，跟 2020 年比。你覺得要真正拿到 CROPS，我們該不該……不要太推那些中心化穩定幣？也許該多推去中心化穩定幣，甚至 ETH 這個代幣本身。我覺得，先講 ETH 本身，對，該有更多人用 ETH。Can I ask a second question? Okay, my second question is about, like, stablecoins. So now we know CROPS and AI are important, but— Stablecoins are maybe the greatest, like, innovation or adoption in crypto in the past 10 years, but it's mostly USDT and USDC. So it's not the way we really, like, it's not trustless ETH. Like, now it looks like we don't say ETH is money anymore, like compared to 2020. So do you think, like, to really get CROPS, should we... Not push too much about all the centralized stablecoins? Maybe we should push more about the decentralized stablecoins or even the ETH, the token itself. Yeah, I mean, I think, like, first of all, about the ETH token itself, I think yes, more people should use ETH, right? I think.

33:33穩定幣與個人化穩定價值

33:33有很多不同情境裡，就算幾個月尺度上有 10% 或 20% 波動，其實完全撐得住。我覺得多數事情就該那樣付。我們做的多數事情就該用 ETH 付。人們想要穩定幣的大原因，我覺得有兩類。一類是人們記帳需要美元。稅務比較單純、會計比較單純，很多事只要用美元計價就比較單純。問題是，你若想做去中心化、以美元為基礎的穩定幣，其實很難競爭。There's a lot of different situations in which, you know, like even 10 or 20 percent volatility on the order of a few months is, like, actually completely survivable. Like, I think most things should just be paid like that. Most things that we do should just be paid in ETH, right? I think the big reason why, you know, people would want stablecoins, I think there's, like, two categories of reasons, right? One of them is people need dollars for accounting, right? And, like, often tax issues are simpler, accounting issues are simpler, a lot of things are simpler if you just charge everything in dollars. And the problem is that it's actually hard to compete if you try to make a decentralized dollar-based stablecoin, right?

34:19原因之一是，只要以美元為基礎，它本來就不是完全去中心化，因為美國政府隨時可以選擇讓它貶值。現在他們連湖都改名了，也許也會開始改美元的價值。另一個問題當然是，去中心化穩定幣很難避免負利率的可能。因為想持有的人太多、願意站在另一邊提供 2 倍槓桿的人不夠，你就很難避免均衡時利率低於零的可能。I mean, one of the reasons why is that if it's dollar-based, it's not fully decentralized anyway because, you know, the U.S. government can always choose to devalue it if they want to, right? You know, like, these days, you know, like they rename lakes, maybe they start changing dollar values too. The other problem, of course, is that, like, with a decentralized stablecoin, the problem is that you can't really avoid the possibility of having a negative interest rate, right? Because if there's too many people who want to hold it and not enough people who want to kind of be on the other side and provide the 2x leverage, then, like, you can't really avoid having a possibility that, like, in equilibrium, the interest rate is below zero, right?

35:10於是問題就變成：價值必須浮動，記帳就太難。所以我其實覺得我們該去弄清楚：不要先做去中心化穩定幣，人們要的到底是什麼，我們能不能用更好的方式提供。我覺得人們要的是去中心化的穩定價值。而我覺得我們能提供、還更好的，是針對每個人個人化的去中心化穩定價值。對你穩定的價值，對住紐約的人穩定的價值，對住……的農夫穩定的價值。And so then the problem is, well, then, like, the value has to be floating, and that makes accounting too difficult, right? And so I actually think we should kind of go and figure out, well, instead of a decentralized stablecoin, like, what is the thing that people want, and can we provide it in a way that's even better, right? And I think the thing that people want is decentralized stable value. And the thing that I think that we can provide that's even better is decentralized stable value that's personalized to each person, right? Like, the value that's stable for you, and the value that's stable for someone who lives in New York, and the stable value for a farmer who lives...

36:00鄉下某處，以及對一個主要支出是算力跟電力的生意來說，穩定價值是什麼？那些是不一樣的。所以我覺得有機會繼續做 DeFi，繼續把它擴成這種更完備的市場：穩定幣、永續合約、預測市場，這些東西某種程度上合在一起。然後每個人基本上能為自己預期的未來支出預付。如果我們做出那個，我們做出的東西對很多人來說大概會太難用。like somewhere in the countryside, and what is stable value for someone running a business whose main expense is, like, compute and electricity? Those are different, right? And so, like, I think there is an opportunity to try to continue working on DeFi, continue to, like, expand it into this kind of more complete market, right, where essentially stablecoins and you know, like, perpetuals and prediction markets and, like, all of these things kind of merge into one, right? And then each person will be able to, like, essentially prepay for their own expected future expenses, right? And if we create that, then we've created something which is probably, like, it might be too difficult for a lot of people to use.

36:48開源硬體還有一關

36:48但對有些人來說，它會是比美元更好的產品。如果我們能做出那個，我們就做出了既真正新、又真正有意思的東西。我們甚至可能是最合適的提供者，尤其如果我們更擅長做出真正安全、真正去中心化的預言機。還有最後一個問題。請。像今天我們談隱私跟安全。But there will be some people for whom it's a superior product to the dollar, right? And I think if we can create that, then we will have created something that's both actually new and actually interesting. And we might even be the best ones to provide it, especially if we become better at making properly secure and properly decentralized oracles, you know? We have one last question. Go ahead. Like, today we talk about privacy and security.

37:34但我看到一個大問題是硬體。現在我們沒辦法控制硬體、知道它在做什麼。甚至有些元件看起來像後門，像 Intel ME。我們在這上面建軟體、建所有東西，挺可怕的。連上面的基礎設施也是。對，我完全同意。我們其實已經開始做很多事，去真正改善開源硬體、開源、以及硬體可驗證性。所以，預期那邊還會有更多我們會貢獻的事發生。But then I see a big problem is the hardware. Like, right now, like, we do not able to control the hardware and know what it's doing. And even, like, some component, it look like the backdoor, like the Intel ME, and it's quite scary when we build the software and everything around. And even the infrastructure on top of that. Yeah, no, I fully agree. And we're actually already starting to do a lot of things to really improve kind of open hardware, open source, and hardware verifiability. So, I mean, expect a lot more things happening on that side that we'll be contributing to.

38:22謝謝，也謝謝 Vitalik。謝謝。Thank you, and thank you, Vitalik. Thank you, sir.