Illia、Jesse 共談 AI 代理經濟與區塊鏈新機遇
The Agent Economy: Why AI Needs Crypto
- Illia Polosukhin NEAR Protocol 共同創辦人
- Jesse Pollak Base 負責人
- Tarun Chitra Gauntlet 共同創辦人暨執行長，主持
場次摘要
AI 與加密產業深度融合，推動資料主權與產品創新。企業與消費者將受益於新基礎建設與平台化趨勢，GPU 市場與供應鏈成未來焦點。
值得記下的 5 件事
逐字稿
三年來的對話與新樂觀
好，我要說，其實我和 Illia 已經連續三年參加這場座談了，這應該是我第一次真的對這個問題的答案感到樂觀。不過，今天我們有兩位很棒的來賓，我想他們對 AI 有完全不同但互補的觀點，等等大家就會知道。你們兩位在加密領域都有很長的歷史，從推出協議到在大型交易所打造 rollup 技術，但不是每個人都知道你們怎麼看加密和 AI，以及你們怎麼參與這些領域。或許可以先請你們簡單介紹一下自己在加密和 AI 的經歷。Alright, I'll say, actually Illia and I have probably done this panel for three years in a row, and this is the first time I actually feel optimistic about answering this question. But, you know, we have two great guests today, and I imagine they have totally different but complementary perspectives on AI, and we'll find out more about that soon. Both of you have a long history in crypto, from launching protocols to building rollup technology at a major exchange, but not everyone knows your views on crypto and AI, and how you've engaged with these fields. Maybe you could each briefly introduce your background in crypto and AI.
Illia 的 AI 經歷比我早開始，應該這麼說。我在進入加密之前，其實是在 Google Research 做 Transformer 架構的 AI 研究，然後 2017 年創辦 NEAR，當時其實是 AI 公司。我們那時候有點太早想要讓軟體自動寫程式，後來發現其實需要一個區塊鏈協議，才能協調人和資料，還有其他東西，而且要去中心化、全球可用。於是 2018 年我們開始做 NEAR Protocol，但過去兩年我主要在思考怎麼讓 AI 保持隱私、可驗證，防止洩密等等。同時，也在想怎麼用區塊鏈經濟市場來啟動我現在稱為單人模式的 AI。Illia's experience in AI started before mine, I should say. Before getting into crypto, I was working on Transformer architecture in AI research at Google Research, and then in 2017 I started NEAR as an AI company. We were a bit too early trying to get software to write code automatically, and then realized we actually needed a blockchain protocol to coordinate people and data, and other things, in a decentralized, globally accessible way. So in 2018 we started NEAR Protocol, but in the past two years I've mainly been focused on how to keep AI private, verifiable, prevent secret leaks, and so on. At the same time, I'm thinking about how to use blockchain market economies to launch what I now call single-player mode AI.
所以，怎麼用區塊鏈當 AI 的後端基礎設施，開啟多人模式？我是 Jesse。我的 AI 經歷沒 Illia 那麼厲害，但我親眼見證了 Base 作為 AI agent 平台的成長，這些 agent 會自動寫程式、買推論、做支付。過去兩年看到這個發展真的很酷，個人來說，我也很熱衷於五年沒寫程式後，現在又能寫 code，這是我今年最開心的事之一。感覺我們真的跨過了一個鴻溝，一年前我還在斷斷續續寫 code，現在我每天會議之間都在寫，很多人應該都有這種體驗。我們正走到一個交會點，會有些事情發生，其實已經在發生了，但我們還不確定會怎麼演變。最棒的是，AI 和加密會以很深刻的方式結合，讓世界變得更好，這讓我很興奮。So, how do we use blockchain as the backend infrastructure for AI to unlock multiplayer mode? I'm Jesse. My AI background is less impressive than Illia's, but I've witnessed Base grow as a platform for AI agents, where agents code themselves, buy inferences, and make payments. It's been amazing to see this develop over the past two years, and personally, I'm really passionate about being able to write code again after five years away—it's honestly one of my favorite things this year. It feels like we've really crossed a gap; a year ago I was trying to code and it was sporadic, now I'm coding all day between meetings, and I think a lot of us have had that experience. We're reaching a convergence point where something is going to happen—it's already happening, but we're not sure exactly how it will play out. The beauty is that AI and crypto will combine in a really deep way to make things better for the world, and that excites me.
AI 與加密的結合契機
我記得 2018 年第一次遇到 Illia，他說他有一家寫軟體的公司，那時還叫 Vibe Coding，大家知道那是黑暗時代的名字。你們剛剛都聊到 AI 和加密，以及你們覺得兩者可能會怎麼互補。歷史上大家其實都蠻悲觀的，可能根本不相信這種結合。我和你（Illia）自 2023 年就一直參加這個 panel，當時大家普遍不太正面，但現在氣氛真的變了很多。那我們先聊聊，你現在對 AI 和加密結合最樂觀的是什麼？I remember meeting Illia for the first time in 2018, and he said he had a software company, which was called Vibe Coding back then—everyone knows that was a dark ages name. Both of you just talked a bit about AI and crypto, and the different ways you think they might be complementary. Historically, people have been pretty pessimistic, maybe not even believing in this kind of combination. You and I (Illia) have been on this panel since 2023, and back then people generally weren't positive, but now the atmosphere has really changed a lot. So let's start—what are you most optimistic about now regarding the convergence of AI and crypto?
我現在聽到最多的問題是：怎麼讓 AI 全球可用？怎麼確保你用的 AI 真的是你想要的，而不是有人在你的 agent 裡塞了有害的東西？這些現在都做得到，像 Near AI、還有其他供應商，當然還有 Venus 等等，你現在就能用，甚至可以用 Zcash、比特幣或其他加密貨幣付款。我真的很期待明年會發生什麼。舉個例子，大家常說你的 agent 以後可以幫你買菜，但其實消費者應用體驗已經很成熟，變化不大。真正體驗很差的是 B2B 商業流程，像找供應商、開發票、審查合約這些，因為這些場景本來就缺乏信任。這正是 AI 和區塊鏈真正結合的地方。The question I hear most now is: how do we make AI globally accessible? How do you ensure the AI you're using is really what you want, and not someone injecting harmful outputs into your agent? This is possible now—like with Near AI and other providers, and of course Venus and others, you can use them now and even pay with Zcash, Bitcoin, or other cryptocurrencies. What I'm really looking forward to next year is, for example, people often say your agent will be able to buy groceries for you, but actually, consumer apps are already so optimized that the experience doesn't change much. The real subpar experience is in B2B business processes, like finding suppliers, invoicing, reviewing legal contracts, because these scenarios inherently lack trust. That's where AI and blockchain really come together.
區塊鏈能提供像託管、支付、金融衍生品等可信基礎設施，AI 則能處理所有繁瑣的事，從 email 溝通到產品搜尋、流程自動化。我認為這樣會讓數兆美元的實體經濟真正搬到區塊鏈上，不再只是投機資產，而是像航運、鋼鐵貿易、化工供應鏈這些真實商業。這些其實已經開始發生，AI 和加密的進入正在推動這一切。Blockchain can provide trusted infrastructure like escrow, payments, financial derivatives, and so on, while AI can handle all the tedious stuff, from email correspondence to product discovery and process automation. I think this will move trillions of dollars of real-world economy onto blockchains—not just as speculative assets, but as real businesses like shipping, steel trading, chemical supply chains. This is already starting to happen, and the entry of AI and crypto is driving all of it.
消費者金融與 AI agent 的未來
我也覺得這一定會發生，尤其是在消費金融和 AI 的交會點。我最近有個親身經歷，幾週前看到有人在 X 上分享一個叫 Beancount 的開源記帳軟體。我就跟我的 AI agent 說，把我所有帳戶都整合進這個帳本，結果它真的幫我整理出完整的帳本、損益表、所有資金流動。這對我來說是個巨大轉折，從完全搞不清楚自己財務狀況，到現在一目了然。我覺得對很多人來說，理財一直很困難，他們不知道怎麼投資、怎麼做預算，常常覺得需要去讀書或上課。現在我們有這些超強的智能 agent，可以整合進我們的財務生活，幫我們管錢、甚至自動交易，讓理財變得更簡單。這讓我非常興奮，我覺得我們才剛開始探索這些可能性。I also think this is definitely going to happen, especially at the intersection of consumer finance and AI. I recently had a personal experience—a few weeks ago I saw someone on X sharing an open-source accounting software called Beancount. I told my AI agent to integrate all my accounts into this ledger, and it actually organized my full ledger, profit and loss statement, and all my cash flows. That was a huge turning point for me, going from having no clear view of my finances to having complete clarity. I think for many people, money is confusing—they don't know how to invest or budget, and often feel like they need to read books or go to school. Now we have these super-powerful smart agents that can be integrated into our financial lives to manage our money, even trade for us, and make personal finance much simpler. This is really exciting to me, and I feel like we're just beginning to explore these possibilities.
從更哲學的角度來看，我覺得我們正經歷 AI 生成內容大爆炸，大家開始質疑什麼才是真的。這時候加密就像一層驗證機制，提供了堅實的基礎設施，剛好能平衡 AI 帶來的這些變化。未來兩年，AI 和加密的結合會以非常深刻的方式重塑我們的世界。From a more philosophical perspective, I feel like we're experiencing an explosion of AI-generated content, and people are starting to question what's real. That's where crypto comes in as a verification layer, providing a solid infrastructure that perfectly balances all the changes AI brings. Over the next two years, the convergence of AI and crypto will reshape our world in a very profound way.
AI 多人模式與區塊鏈經濟
我想說的是，AI 領域的人通常都在做單人模式，但我們這些做加密的，早就習慣了多人競爭和激勵對齊。把這兩種模式結合起來，才是解方。我寫過很多關於怎麼把這些原則應用到 AI 推論市場和算力市場的文章。其實這正是 AI 思維裡常常缺少的一塊。現在很多消費者應用都在做 agent security protocol，讀他們的文件你會發現：等等，你們其實是在做假的區塊鏈。AI 專家其實在不知不覺中重新發明區塊鏈，因為他們討厭 blockchain 這個詞，覺得它是個髒字。What I'd say is, people in AI usually work in single-player mode, but those of us in crypto have long been used to multiplayer competition and incentive alignment. Combining these two modes is the solution. I've written a lot about applying these same principles to AI inference markets and compute markets. This is actually the missing piece in a lot of AI thinking. Now, many consumer applications are building agent security protocols, and when you read their docs, you realize: wait, you just built a fake blockchain. AI experts are actually reinventing blockchain concepts without realizing it, because they hate the word blockchain—it sounds like a dirty word to them.
我們來快速回顧一下過去幾年加密和 AI 的發展。最早大家都在瘋狂模仿比特幣，像是出租 GPU，說我可以給你一個節點，這裡有個去中心化節點網路。其實加密圈很早就看對了市場，現在 AI 推論供應商一年有幾十到幾百億美元的市場，但加個代幣並沒有讓產品變得更好。後來大家又轉去做去中心化訓練或 AI agent，其實像 AXBT 這些 AI 交易 agent 反而太早了。你怎麼看？回頭看，很多加密創業者其實想做 AI，最後都去創了 AI 公司，像 Alexey 去做 OpenRouter，他說我們不需要加密貨幣，可以接受但不需要區塊鏈。其實很多礦工像 CoreWeave 也都轉去做 AI，現在一半新雲端都是礦工出身。你會發現很多基礎設施都這樣，大家說我們要做這個，但又覺得區塊鏈反而拖慢進度，這背後有幾個原因。大家常說代幣可以激勵網路效應，但過去兩年其實沒那麼有效。新代幣模式沒有想像中成功，因為代幣本身就是一個產品，你同時要做代幣和 AI 產品，這其實很難，因為 AI 產品本來就很難做。Let's quickly look back at the past few years of crypto and AI development. At first, everyone was obsessed with mimicking Bitcoin, like renting out GPUs and saying, I can give you a node, here's a decentralized network of nodes. Actually, the crypto community saw the market correctly early on—now AI inference providers have tens to hundreds of billions of dollars in annual market size, but adding a token didn't make the product better. Later, people moved to decentralized training or AI agents, but actually, things like AXBT—AI trading agents—were too early. What's your take? Looking back, a lot of crypto founders who wanted to do AI ended up starting AI companies, like Alexey who started OpenRouter—he said we don't need crypto, we can accept it but don't need blockchain. In fact, many miners like CoreWeave have shifted to AI, and now half of new clouds are run by former miners. You'll see this across infrastructure—people say we need to do this, but then feel blockchain slows them down, and there are a few reasons for that. People often say tokens are great for bootstrapping network effects, but in the past two years that hasn't really worked. The new token models haven't been as successful as hoped, because the token itself is a product—you're trying to build a token and an AI product at the same time, and building good AI products is already hard.
產品與代幣啟動的時機抉擇
所以我認為這也跟注意力分散有關。我現在覺得有些公司會在產品建立後才開始用代幣，這樣可以啟動網路效應。So I think it's also about distraction. I think now some companies actually use a token after they've established their product, and they can use it to start network effects.
我想說市場正在改變。有些情況真的就是時機問題，像是先發代幣再試著做產品，像 AI16Z 那樣，這其實是很糟的順序，對吧？他們本來可以先把產品做好，再發代幣來激勵經濟。I'd say the market is changing. I think some of this is literally about timing, where you launch a token first and then try to build a product, like AI16Z and so on. That's a really bad sequence, right? They could have just built the product first and then launched a token to incentivize the economy.
我覺得這歷來就是個挑戰。沒錯，你剛剛提到的那點其實是全球通則。很難在產品還很早期時就發幣，然後還能成功。I think that has historically been one of the challenges. Yeah, and I think the last point you mentioned is kind of a global rule. It's very hard to have a product at an early stage and launch a token very early and make it work.
因為你得不斷調整產品，同時還要管理這個經濟系統。我覺得這又多了一層難度。有些人能做到，像我看到 Venus 就是，他們在生命週期還算早的時候就發幣，但 Eric 帶領 Venus 已經有十年經驗了。Because you have to do a lot of iteration on the product, and at the same time you have to manage this economic system. I think that adds another layer of difficulty. There are some people who can do it, like I see with Venus. They launched it relatively early in their lifecycle, but Eric leads Venus and has a decade of experience navigating those waters.
時機與產業成熟度的關鍵
我不認為大家真的做錯了什麼，也不覺得過去的時期都是錯誤。我覺得很大一部分是時機問題。這些事本來就需要很長時間。如果你看過去五年加密系統的成熟度，現在我們有穩定幣、有股票代幣化，還有其他資產，鏈變得又快又便宜又好用。I don't necessarily think people made mistakes or that the past periods were just mistakes. I think it's largely about timing. These things take a long time. If you look at how mature crypto systems have gotten in the last five years, we now have stablecoins, we have tokenized equities, and all these other assets. Chains have become fast, cheap, and easy to use.
這些基礎建設以前根本不存在。四、五年前，AI 這邊的模型也還沒成熟到可以真正補足加密系統。所以，理論上加密圈的人能不能做出 OpenAI 或 Anthropic？當然可以，但我預期這兩邊會逐漸融合，等到技術都成熟。This infrastructure didn't exist. Four or five years ago, on the AI side, the models weren't at the stage where I think they were mature enough to really complement crypto systems. So, could crypto people have invented OpenAI or Anthropic? Sure, but my expectation is that they're going to converge as the technology matures.
未來幾年我們會看到越來越多整合。舉例來說，Venus 就是 AI 結合加密貨幣的成功案例。它對公司、資產、AI 代理人來說都有效。我很樂觀這個循環會繼續下去，大家會不斷找到交集。And over the next few years, we'll start to see more and more integration between them. That's true, right? If you look at Venus, it's a private AI with a user experience that has crypto built in and is working successfully. It's working for the company, it's working for the asset, and I think it's working from the agent and AI perspective. So I'm very optimistic that this flywheel will keep spinning, and people will keep finding those intersections.
開放模型與產品導向的成功關鍵
你提到的這些成功案例真的很棒，尤其是今年。像 Venus 這種產品，其實也是因為開放模型的進步。開放模型的生態系在原則上跟加密貨幣很像，發展歷程也很相似。那你覺得大家做對了什麼？為什麼過去十二個月 AI 和加密專案能有這麼大突破？像 Nouse 的 Hermes、NEAR、Venus 等等，有些有發幣，有些沒有，但起點都一樣。你覺得最關鍵的因素是什麼？未來十二個月你預期會怎麼發展？Yeah, I think you make a great point talking about some of the successes we've seen, especially recently this year, right? I think a product like Venus is also a result of better open models. The open model ecosystem is much closer in its principles to crypto. Its origin looks very similar. So what do you think people got right? And why has there been this big shift in the last twelve months where crypto and AI projects have achieved broader success, starting from Nouse with Hermes, NEAR, Venus, and others? Some had tokens, some didn't, but they all started from the same point. What do you think are the biggest factors for these projects' success, and where do you expect things to go in the next twelve months so that next time I'm here with Illia, we have a new answer?
好產品，這好像是唯一不變的道理。我自己用 Hermes agent 已經快半年了，很喜歡。我也用 Venus，真的很棒。NEAR 在意圖面也找到產品市場契合。所以，crypto 作為經濟模型、激勵工具、融資方式，絕對是超能力，但你還是需要好產品。我覺得我們產業已經過了關鍵點，現在大家都很專注在這件事，這也成了評價標準，自然就有更多成功案例。A great product, I don't know. That seems to be the constant. I've used my Hermes agent for almost six months, and I love it. I use Venus, it's a great product. Obviously, NEAR has found product-market fit on the intents side. So again, crypto as an economic model, as an incentive tool, as a way to fund things, I think it's a superpower for sure, but you still need great products. And I feel like as an industry, we've generally crossed a critical point where we're very focused on that. That's the standard everyone is judged by, and that means we're naturally starting to see more successes.
開源模型進步與 AI 代理人普及
我同意你的看法。產品導向很重要，同時也要搭上開源模型進步的浪潮。現在開源模型已經非常接近 frontier models，甚至整體水準也很高。我們用 DeepSeek Flash 做正式驗證，表現超越所有人，包括 GPT、Claude，還有那些花幾億美金訓練的專屬模型。所以現在模型真的很強、很便宜，也能跑更久。這讓你可以運行 agent，也能做私有運算，現在真的能看到實際成果。Hermes 也帶動了個人 agent 這波潮流。其實跟所有新創一樣，定義一個新類別就是你做出好產品，然後搭上某個浪潮，把產品帶給越來越多用戶。Yeah, I agree with you. I think the focus on product, and at the same time riding the wave of what's happened, is that open-weight models have gotten so much better, right? They're much closer to frontier models, but also generally at the level now. The example we use is that we run DeepSeek Flash for formal verification and outperform everyone, including GPT, Claude, and custom models that people spent hundreds of millions of dollars to train. So the models are actually very good, they're cheap, and you can run them longer because of that. And so you can run agents, you can run private compute, and you're actually getting real results now. So I think that was a big move. Obviously, Hermes also kicked off this personal agent movement. So I think, as usual with any startup, the definition of the category is that you build a great product, you catch a wave that you can ride by being there, and you deliver that really great product to an expanding audience.
加密創業者與 AI 融合的優勢
我覺得加密創業者在整合 AI 產品各層面上做得更好。也許一年前開源生態還沒那麼成熟，從消費產品到模型效能的轉換能力還不夠，沒辦法真正發揮效能。但現在這些模型已經進入大家生活，大家也開始更重視隱私，這正是加密貨幣可以介入的地方。你怎麼看私有運算、私有推論、私有 AI 的需求發展？這跟加密貨幣會怎麼交集？Yeah, I think one thing is that crypto founders have been much better at being able to integrate different aspects of AI products, whereas maybe a year ago the open-source ecosystem wasn't very good. The ability to go from a consumer product to model performance and harness that performance, etc., wasn't available. So I think, looking a bit to the future, as these models become embedded in people's lives, people are starting to value privacy much more, right? And that's where you start to see more intersection with crypto. So I'm curious, how do you see the evolution of demand for private compute, private inference, private AI? And where does that intersect with crypto?
私有 AI 與數據主權的市場轉變
我們去年十二月推出私有解決方案，剛開始市場還沒準備好，大家都用主流模型，企業也只是試試看，沒那麼急迫。但三、四個月前情況完全變了。現在大家會把公司所有資料都丟給 AI 模型，任何有權限的人都能重建公司全貌。這些模型會做關鍵決策，所以大家很想知道到底發生什麼事，想要能審查模型、知道用的是哪個模型。不是說第一週很棒，第二週就變笨了，大家不想這樣。消費者也一樣，現在越來越多人把財務、醫療等敏感資料交給 AI，這些決策都在那裡，所以很重要不能亂流出去。Yeah, I mean, I'll say we launched our private offering in December of last year, and when we started offering and selling it, the market wasn't ready, right? A lot of people were using frontier models, so you go to companies and they say, 'Yeah, we're fine, we're experimenting, it's not critical.' And I think even in the last three or four months, the situation has completely changed. People now say, 'Okay, I'm actually giving these AI models my entire company data.' So anyone with access to these logs can literally reconstruct the company from them. They're making critical decisions, so I really want to know what's going on, right? I want a way to audit the model and know the exact model, right? Not like, you know, week one was great, week two got a lot dumber, right? So I don't want that to happen. And even from a consumer perspective, right? People are now giving more and more financial data, medical data, all these decisions are there. So it's important that this doesn't go somewhere else.
每次 OpenAI 爆出訓練用到不該用的資料，或像 Anthropic 跟 Figma 合作後推出 Claude Design，這些都讓人更想要掌控自己的資料。這就是加密的哲學核心——主權、控制、隱私、可驗證、可存取。市場已經大幅轉向這個方向。And obviously, every time there's news about OpenAI, that maybe they trained on some data they weren't supposed to and used it to influence stocks, or the Anthropic example with launching Claude Design after being partners with Figma, all of that doesn't help at all. And obviously, them coming out and saying they're going to wipe everyone out, you know, there's a 10% chance, so they're kind of undermining their own position by doing that. So people increasingly want to say, 'Hey, we really want to be in control.' And that's literally the core of crypto from a philosophical perspective. It's about sovereignty and control, getting privacy, verifiability, and accessibility. So yeah, I think there's been a huge shift toward this in everything happening in the market.
企業與消費者的 AI 主權實踐
我們在 Coinbase 內部也直接體會到這一點。Coinbase 很大，我們今年初還完全依賴主流模型，大家都用同一家供應商。後來 Brianna 在第一季末說：我們不能只靠一個模型或供應商，要學會用開源模型，掌握自己的資料。過去六個月我們大幅多元化，現在有自己的系統，也用一些主流模型，還有很多開源模型。我們現在更清楚資料流向。整體 token 消耗量在增加，但成本基本上下降且穩定。AI 讓團隊生產力大幅提升，比六、九個月前好很多。我覺得這個趨勢會蔓延到所有企業，開源、隱私、控制、成本管理都會成為主流。消費者也會跟上，雖然可能慢一點，但這趨勢很明顯。大家都想要主權，這也是我們在這產業的原因。加密社群在這方面走得比較前面，但我預期會普及到各行各業。Yeah, and I think we've seen that directly internally at Coinbase, where if you look at our usage trajectory, you know Coinbase is a big company. I think we pushed hard toward adopting AI in general. I can say that at the beginning of the year, we were entirely dependent on frontier models. We were completely locked into a single provider that everyone was basically using. Then, thanks to Brianna at the end of Q1 this year, she simply said: we need to not be locked into a single model or provider. We need to figure out how to use open-source models and get to a place where we control our data. And over the last six months, what happened is we've diversified a lot. We're now running our own systems, distributing work to some frontier models, plus a large number of open-source models available to us, and we're much more aware now of where all the data is flowing in the business. And if you look at total token consumption, it's going up, but if you look at costs, we've basically been able to get them down and stable. And I think from a productivity perspective, we've made the team much more productive and successful with AI than we were six or nine months ago. So my sense is that this same trajectory is probably going to hit every business. It's going to come for open source, it's going to come for privacy, it's going to come for control and cost management. I think it'll also come to consumers, but they'll probably be a little bit behind, although that's happening pretty clearly. But yeah, people want sovereignty. That's why we're all in this space. I think the crypto community is a little ahead of the curve, but I expect it to spread everywhere.
加密創業者的技術整合與平台化
我想補充你剛剛說的，關於加密創業者在不同技術層面整合的能力。像 Venus 的 DM 模型，他們先有消費產品，再思考怎麼把推論 token 化、上鏈，然後開始打造市場。我覺得這就是很好的例子，證明先有好產品，再用加密技術讓產品變成平台，更可組合，甚至創造新市場，讓大家對自己的推論有更多掌控權，也提升我們對 AI 使用的主權。希望未來幾年 crypto 和 AI 的發展，我們這些堅持的人能繼續發聲，不讓少數實驗室掌控未來。There's something I wanted to separate from the privacy point, but I wanted to comment on what you said about crypto founders being uniquely good at knowing how to work across different levels of technology. One thing that excites me is part of the Venus product suite, what they did with the DM model, where they said at the core: we have this consumer product, let's figure out how to tokenize inference and put it onchain, then start building a market around it. And I think that's a great example of knowing, okay, we've built a great product, now how do we leverage crypto tech to make this product more of a platform, make it more composable, and honestly, start inventing new markets that increase access, give people more control over their core inferences, and again, increase our sovereignty over how we all use AI. So my hope for how crypto and AI evolve over the next few years is that we are the people who stick around to say, yes, we're not going to let just a few labs control the future.
資料主權與AI經濟新契機
我們將擁有開放的經濟、開放的原始碼、開放的系統，這就是我們要打造的未來。We will have open economies, we will have open source, we will have open systems, and that's the future we're going to build.
這一切都會建立在這個新平台之上。All of this will be built on top of this new platform.
其實我想回到推論市場的成長，但這讓我想到一個問題，很多加密圈的人其實很早就注意到一些事情。Actually, I want to go back a bit to the inference market and how this has grown, but it really made me think of one question, where you know, there are a lot of things that crypto people were really early on.
像是2019到2022年Web3運動期間，有一個很大的重點就是擁有自己的資料，對吧？你可以因為自己的資料而獲得補償，所有的東西都朝這個方向發展。One thing about the Web3 movement between 2019 and 2022, there was a big focus on owning your own data, right? In the sense that you get compensated for your own data, and everything you contribute goes in that direction.
但在現在的AI世界裡，每個領先的實驗室，甚至開源公司，都在挖掘你的足跡來優化他們的模型和工具，卻沒有真正的補償。But in the current AI world, every leading lab, even open source companies, are digging into your traces to improve their models and tools, and there's no real compensation.
某種程度上，Web3最初關於資料擁有權、來源和主權的想法，其實應該要在這裡落實。In a way, a lot of the original Web3 ideas about data ownership, provenance, and sovereignty should really be applied here.
你覺得這幾年內會發生嗎？還是說這還很遙遠？加密圈是不是早了十年，而不是幾年？So, do you think that's going to happen in the next few years? Or do you think it's still far off? Were crypto people a decade early instead of just a few years?
其實我們現在有個產品正在開發中，先賣個關子，但它真的可以讓你獲得私有推論。如果你願意，也可以貢獻你的資料足跡，換取更多推論或其他方式的報酬。Actually, we have a product in development right now. We have to tease it, but it actually lets you get private inference, and if you want, you can contribute your data traces and get compensated either in the form of additional inferences or in other ways.
這些資料之後會被用來訓練模型。有趣的是，怎麼知道這些資料的真正價值？因為我們過去談資料擁有權時，最大的問題就是沒辦法好好評價資料的價值。And effectively, that data will later be used to train models. The interesting thing is how to know the real value of that data, because one of the problems we faced when talking about owning your data before is that we didn't find a good way to value how much that data is worth.
有時候看起來就像一堆隨機的文章或文件，內容可能毫無價值，也可能是Coinbase內部的高價值文件，根本不該被分享。It just seems like a random collection of articles or documents. The content could range from worthless to highly valuable internal Coinbase documents that shouldn't be shared.
但現在有了代理人導向的資料，評價方式變得更清楚，因為有人類輸入來修正模型，這非常有價值，也可以用程式化方式評估。But now with agent-driven data, there's a clearer way to value it, because you have human input to correct the model's trajectory. That's extremely valuable and can be evaluated programmatically.
我認為這個發展會出現，但必須結合可驗證的推論。你知道AI的輸出真的來自來源，所以你需要打造一個平台，讓大家可以參與資料評價和定價。So I think we'll see this development, but it needs to be combined with verifiable inference. You know the AI output actually comes from the source, so you need to build the platform to reach that level of contribution in data valuation and pricing.
資料變現與企業主導的轉型
我認為成本會由實驗室負擔，不一定會從消費者端開始。有個有趣的趨勢是我們跟AWS和Cloudflare合作，他們都發現有很多企業在網路上經營業務，卻一直被實驗室和代理人挖資料，損失獲利機會，也不知道怎麼處理。I think the costs will be borne by the labs. I'm not sure it'll start from the consumer side. One interesting trend we're seeing is we've worked with AWS and Cloudflare, and both independently pointed out that they have many companies running businesses on the web and constantly getting their data scraped by labs and agents, losing profit opportunities, and they don't know how to deal with it.
他們想到的解決方案是讓這些企業可以對每一頁內容收費。我們跟他們合作，讓這件事變得簡單，如果你用Cloudflare或AWS架站，可以直接開啟選項，設定每頁價格，可能是每次載入收千分之一或百萬分之一美分，透過XRP2支付，最後在Base網路用USDC結算。The solution they came up with is to enable these companies to charge per page for that content. So we worked with them to make it simple: if you run a site on Cloudflare or AWS, you can enable an option and set pricing per page, maybe a fraction of a cent per load, paid via XRP2, and settled in USDC on the Base network.
這樣新聞網站就能對每次被代理人或實驗室挖資料收取千分之一或百萬分之一美分，這些金額會累積起來。我覺得明年你會看到一個轉變，過去五到十年實驗室能免費取得大量網站和網路資料，現在這些網站會意識到這點，開始要求付費。This way, a news site can charge a thousandth or a millionth of a cent each time it's scraped by agents or labs, and these amounts add up. I feel like what might happen in the next year is you'll see this shift, where labs have been able to get a lot of free data from websites and the internet over the past five to ten years, and now all those sites will realize this and start demanding payment.
同時，原始資料的價值會提升，消費者也會意識到：如果這些公司都因為資料賺錢、股價上漲，為什麼我不能分一杯羹？所以消費者端也會出現同樣的情況。我認為未來會有像你們正在開發的產品，讓用戶可以說：你可以交易我的資料，但必須付錢給我。At the same time, the value of raw data being fed in will increase, and consumers will realize: oh, if all these companies are making money and that's why their stock prices are going up, why can't I get my share? So we'll see the same thing happen on the consumer side. I don't see any reason why there can't be products like what you're building, where you can say: yes, you can trade my data, but you have to pay me for it.
我會這麼做。對，這領域有很多機會。I would do that. Yes, there are a lot of opportunities in this space.
企業主權與消費者體驗的演進
你剛剛點出一個有趣的現象，當我想到Web3時，會想到2001年，那時重點都在消費者個人，像是消費者主權、網路主權。但你們兩個的例子都聚焦在企業主權，感覺企業比消費者更重視自己的主權，這也是Web3在這個市場的某種演化。It's interesting you highlighted something funny—when I think of the Web3 era, I think of 2001. The focus was really on consumer personal stuff, like the sovereign internet for consumers. But in both your examples, you focused on corporate sovereignty. It seems like companies care more about their sovereignty than consumers, which is a kind of evolution of Web3 for this market.
我認為這也是為什麼加密貨幣在AI市場更容易被接受，這反映了產業的現狀。從產品市場適配來看，交易、支付、金融已經在企業間廣泛採用，因為用新基礎建設比舊的更快、更便宜、更好，企業因此轉向新架構，消費者則因為產品變好而受益。I think that's also why crypto aspects are more accepted in the current AI market, and I think that's an indicator of where we are as an industry. If you look at product-market fit, trading, payments, and finance have reached wide adoption among companies, because it's easier, faster, cheaper, and better for them to run their business on this new infrastructure instead of the old one. That drives companies to move, and the consumer still benefits because they get better products.
但現在消費者不需要成為加密專家，企業主可以選擇最適合自己業務的技術，這最終會讓所有客戶受益，這其實是很棒的改變。But now it doesn't require me as a consumer to be a crypto expert. As a business owner, I can choose the best technology to run my business optimally, and that benefits all my customers. It's actually a really great change.
這也讓過去兩年加密圈的情緒和身份有點動盪。我們以前都在賣加密貨幣的概念給大眾，但現在我們賣的是更好的基礎設施，幫助大家打造更好的產品。I think this has definitely led to a bit of disruption in the emotional state of crypto over the past two years. We all used to sell the idea of crypto to people, but now we're not selling that idea anymore, we're selling better infrastructure to help people build better products.
我覺得在Venice也看到這個現象，消費者還是有需求，只是你真的要打造一個符合消費者體驗水準的產品，這也是我們在Nair.com做的事。I mean, I think we see this in Venice too, right? There's still a consumer story. You really need to build a product that matches the level of consumer experience, and that's what we're doing at Nair.com.
這是一個完全抽象化的系統，沒有Gas費、沒有鏈、沒有複雜的東西，只要能運作就好。當達到這個水準，消費者自然會採用，之後你再教育他們隱私和主權的價值。It's a fully abstracted system—no gas fees, no chains, nothing complicated. It just needs to work. When you reach that level, consumer adoption comes, and then you can educate them about the value of privacy and sovereignty.
但加密產業一直很難做到這種體驗，主要是基礎建設和其他挑戰太多。對企業來說，一旦監管鬆綁，他們發現支付成本可以降一百倍，當然就會採用。But I think crypto has struggled to reach that level of experience for the most part, and there have been a lot of infrastructure and other challenges. With companies, it's easier—once regulation is actually lifted, they see that doing payments is a hundred times cheaper, so of course they'll do it.
企業的決策很理性，都是看Excel表格，而不是只看產品體驗。雖然Web3 for企業聽起來有點無聊，甚至有點悲傷，像是企業主權？對，這就是我要說的，也許需要重新包裝一下。It's very calculated for companies, based on spreadsheets rather than always on product experience. But, you know, saying Web3 for companies sounds a bit like boring corporate talk. It's kind of sad, like corporate sovereignty? Yeah, that's what I was going to say. Maybe it needs a bit of rebranding.
AI與加密融合下的新市場機會
剩下兩分鐘。你們都提到因為加密和Web3融合，出現了很多新市場。我最近一直在思考並寫作，像是開放式推論市場加上保有用戶資料的工具，這其實很像加密貨幣，把實驗室解構掉。這些工具儲存用戶資料，就像錢包一樣，導向器決定你該去哪，模型像是公共財，推論提供者就像流動性供應者。We have two minutes left. You both mentioned that new markets have emerged from the fusion of crypto and Web3. I've been thinking and writing a lot about this lately—open inference markets plus tools that retain user data, it really resembles crypto, breaking up the labs. Those tools store user data, like a wallet; the router tells you where to go; models are like public goods; inference providers are like liquidity providers.
現在我們看到類似加密的市場結構開始在AI出現。你們覺得明年會出現什麼新的市場結構？我們有推論、資料、還有其他用戶變現方式。你們分別站在不同的應用場景，覺得AI和加密的下一個大市場會在哪？Now we're seeing crypto-like market structures start to appear in AI. Where do you think the next big new market structure will emerge in the next year? We have inference, data, and other forms of user monetization. As you both sit in different use cases, where do you feel the next big market in crypto and AI will be?
我認為最大的是GPU本身，這個市場現在完全混亂，充滿OTC交易，缺乏結算場所，這是目前最大的機會，很多人都在搶這塊。接下來我最期待的是更廣泛的商業供應鏈，像是企業間的合作，把Web3和AI結合起來，降低所有營運成本。The biggest one is GPUs themselves—the market is completely disrupted right now, full of OTC transactions, lacking a place for settlement. That's the biggest opportunity at the moment, and a lot of people are chasing it. The next thing I'm really excited about is a broader commercial supply chain, like doing business-to-business relationships, leveraging Web3 and AI in one place to reduce all the costs of doing business.
我覺得還有一個跟意圖有關的東西，隨著我們越來越多介面和工作導向系統，無論是推論、資料、交易或執行，這些人類意圖會被標準化，然後由不同服務提供者找到最佳執行方式。I feel like there's also something related to intent. As more and more of our interfaces and work are routed through systems—whether it's intent for inference, data, execution, or trading—these human intents get wrapped in a standard format, and then the best execution is found across a set of different service providers.
我還不確定這個市場會長什麼樣，但我剛剛才跟團隊討論，這其實就是意圖和訂單簿，像是流動性池和市場，有供給端和需求端，最後會以某種代理人形式結合，跟加密貨幣兼容，價值很高，我覺得這幾年就會出現。I'm not sure exactly what that thing will look like or what the market structure will be, but I was just talking to someone on my team—it's about intent and order books. It's like pools and markets: you have where supply is built, where demand aggregates, and those will come together in some agentic, crypto-compatible, high-value form. I think that's likely to emerge in the next few years.
最近很多討論都在講結果導向的定價，不是按token計費，而是按你拿到的結果付費，這其實就是AI的語言。我覺得這一定會來，已經在發生了。There's been a lot of talk recently about outcome-based pricing instead of paying per token—you pay for the result you get, and that's the AI language for the same thing. I think that's inevitably coming. It's already emerging.
時間到了，謝謝大家，希望你們有收穫。謝謝大家。That's time. Thank you all. Hope you got something interesting out of it. Thanks, everyone.
