Ray Xiao

638 posts

Ray Xiao banner
Ray Xiao

Ray Xiao

@_RayXiao

Investing @OKX_Ventures. Curiosity is the North Star. Technology diffusion is as vital as innovation. Prev @IOSGVC Opinions are my own.

Katılım Ekim 2011
480 Takip Edilen1.9K Takipçiler
Sabitlenmiş Tweet
Ray Xiao
Ray Xiao@_RayXiao·
🧵 1/ Beyond the Checkout Page: Who Will Build the Economy for Agentic Commerce? A deep dive into the battle between human-centric design and machine-native protocols. As AI agents evolve from assistants to autonomous entities, they're reshaping commerce. Here is the full link of the research: okx.com/en-sg/learn/be…
OKX Ventures@OKX_Ventures

The future of commerce is here: Agentic Commerce is redefining how we shop, pay, and trust in a machine-driven economy! From AI agents handling purchases to crypto-powered, machine-native payments, we're on the cusp of a revolution. 💸 Read more: okx.com/en-sg/learn/be…

English
6
1
22
4.7K
Ray Xiao retweetledi
mariaa.eth 🐸
mariaa.eth 🐸@MariaShen·
We mapped 501 sources of real-world yield in traditional finance. Only 34 have any meaningful on-chain presence. We break down: - what keeps the other 467 off-chain - why distribution is the biggest bottleneck - what needs to be built to bring them on
mariaa.eth 🐸@MariaShen

x.com/i/article/2034…

English
17
21
122
24.4K
Ray Xiao retweetledi
OKX Ventures
OKX Ventures@OKX_Ventures·
The dominant narrative in RWA has been tokenization — bringing real world assets onchain for faster, cheaper settlement. That’s a genuine unlock. But the far larger engine of global markets remains almost untouched: leveraged directional trading. Enter RWA Perps — the cleanest bridge yet between DeFi and Wall Street 🧵👇 okx.com/en-sg/learn/ok…
English
2
7
421
2.9K
ZY
ZY@yezhang_cn·
不得不说 我的内核还是一个researcher,只不过过去的经历培养了我很多创业者的品质和思维方式,会习惯性的思考PMF,strategy,execution以及更加long term的东西 但是夜深人静坐下来的时候,我发现我最感兴趣的东西还是从第一性原理理解各种复杂的系统,深刻的理解很多问题的本质(这个过程本身就令我非常愉悦),很久以前对ZK也是这样的,是数学构造本身的优美先吸引了我,然后才发现blockchain里面的很多应用 =_= (今天读了一天AI Interpretability的东西,所以真的有感而发) 我感觉很多interpretability的研究能回答我大量的疑惑,也深受启发。我觉得只有打开黑盒 理解这个黑盒 搞清楚这些东西的本质才能让我真正的有conviction。我现在越来越觉得Anthropic的方法是正确的,alignment & safety本质要研究的还是AI的interpretability,怎么去解释很多AI的推理过程和behavior,理解fine tune究竟是在干啥,我觉得这个一定是通向AGI的灯塔和明灯 大胆去想,假设你的脑海里有一个model最first principle本源的样子,你深刻知道自己的每一次train在影响哪些东西的链接,数据在另一个维度究竟在发生什么变化,知道你的input到output的每一个token都是怎么被产生出来的,每一个单词为什么一个个蹦出来还能产生这么专业的相关性。我感觉这个就是神的视角,能做到降维打击。从这个角度而言interpretability才是train model的first principle的本源。 当然也许我还是过于乐观了,现在很多解释还是比较浅层的,也没法解释scaling law之类的很多堆数据起来的东西,不过我觉得一个人如果真的能对这些问题有很好的直觉,才能把算法,infra这些锤子打在钉子上,才能在接下来1-10年更好地生存 研究这些问题的时候,很像回顾了我当年思考和学习数学的approach。我从小到大一直数学都很好,也搞了竞赛,但是我并不觉得我真的是绝顶聪明的奇才,而是我感觉我有get一种很好的intuition,能知道什么东西是重要的 什么东西不是重要的: 比如很多人做不出来题看完答案总是豁然开朗 — “啊 原来是这样做的”,但是我觉得最需要思考的是 — “我在相同的起点 为什么没能想到这个解法”,我觉得其实最难的是 从无到有,真正意义上想到 怎么想到这个解法 的过程,才是真正需要训练的,而不是读完答案以后当个verifier,然后下一次继续选择更高概率的解法,然后兜圈子 突然蹦出来很多想法。也许未来50%以上的概率会有新的方法横空出世,新的公司可以真正的做到降维打击现在的LLM,真正的拥有泛化能力,用很少的数据量自由的探索很多out of distribution的东西,能做到方法的迁移,能够做一个更会思考学习的model。我隐隐能感觉到利用coding类似的approach,verifiability能做出来很多有用的东西,ai会尝试很多东西 会犯错 但是会做很多人类没有探索过的新鲜尝试,并且在数学物理化学生物的范畴内做到可验证,然后在基于这个模型扩散,慢慢进入生活
中文
2
0
4
146
Ray Xiao retweetledi
OKX Ventures
OKX Ventures@OKX_Ventures·
Trade Everything, Always: Stocks, Gold, and Forex on-chain? Let’s talk RWA Perps! Join @OKX_Ventures on March 5th, 12:00 PM EST to explore: ⚡️ RWA Perps vs TradFi (CFDs, 0DTEs) & Altcoin Perps 🌐 Overcoming UX challenges & the 24/7 weekend trading tradeoff 🏛 The massive Web2/Web3 demand & NYSE's 24/7 tokenization impact 🎙️ Featuring insights from: @OstiumLabs @Lighter_xyz @felixprotocol @VestExchange Set your reminder! 👇 x.com/i/spaces/1akem…
OKX Ventures tweet media
English
6
9
41
4K
Ray Xiao retweetledi
Star_OKX
Star_OKX@star_okx·
A New Chapter: Building the Next Generation of Financial Infrastructure Our partnership with Intercontinental Exchange marks an important moment for OKX and for the broader evolution of digital asset markets. ICE has built and operated some of the most important financial infrastructure in the world, including the New York Stock Exchange and global derivatives and clearing platforms. Their decision to invest in OKX, and join our board, reflects a shared belief that digital asset technology will play an enduring role in the future of financial markets. For OKX, this partnership also represents a new chapter in how we approach the United States. In many ways, we view our presence in the U.S. as a blank sheet of paper — an opportunity to build thoughtfully, engage constructively with regulators and institutions, and contribute to the development of market infrastructure that meets the standards of the world’s most sophisticated capital markets. Financial markets are entering a period of structural transformation. Blockchain technology allows assets to move and settle globally with unprecedented efficiency. Artificial intelligence is reshaping how markets analyze information and manage risk. At the same time, expectations around safety, transparency, and investor protection remain as important as ever. The next generation of financial infrastructure must bring these elements together. One area where we see tremendous potential is the development of tokenized securities and digital representations of traditional assets. In the future, issuers may be able to bring securities directly to global investors through modern digital infrastructure, while still benefiting from the governance, market structure, and regulatory frameworks that have long defined traditional exchanges. Working alongside ICE and the broader New York Stock Exchange ecosystem gives us a unique opportunity to explore how these models can evolve responsibly. Our focus is not simply on new technology, but on building durable infrastructure for the global financial system. This includes improving market structure, strengthening risk management and clearing frameworks, expanding institutional access to digital assets, and creating platforms that protect consumers while enabling innovation. OKX today serves more than 120 million people globally and operates under licensing frameworks in major financial jurisdictions. Over the past decade, we have built high-performance trading systems, onchain technologies, payment systems and security frameworks capable of supporting large-scale global markets. As digital assets continue to mature, we believe collaboration between technology innovators and established financial institutions will be essential. Our partnership with ICE reflects this principle. Together we will explore how traditional exchange infrastructure and digital asset technology can complement each other to build stronger, more efficient markets. This investment is not an endpoint — it is the beginning of a deeper collaboration. Our goal is to help shape the next chapter of financial markets, where digital and traditional infrastructure work together to expand access, strengthen trust, and support innovation across the global economy. okx.com/en-us/learn/ok…
English
587
279
1.6K
929K
Ray Xiao retweetledi
OKX
OKX@okx·
AI agents onchain need reliable rails. That's why we’ve added a native AI layer to OnchainOS - the first dev toolkit enabling agents to manage wallets, make payments, trade & read markets. Backed by the same infra trusted by 12M+ monthly @wallet users. Learn more: okx.com/learn/onchaino…
English
152
274
2K
12.7M
Ray Xiao retweetledi
jolestar
jolestar@jolestar·
MCP vs CLI 不是对立问题:真正缺的是调用层 最近刷到几篇关于 MCP 和 CLI 关系的文章。 看完后的感觉是:把这两个东西放在“二选一”的框架里讨论,本身就是个错位。MCP 和 CLI 根本不在一个层面上。 MCP(Model Context Protocol)解决的是能力如何标准化暴露。 CLI 解决的是能力如何被调用。 一个是能力面。 一个是调用面。 现在的问题不是 MCP 能力暴露这层出了问题,而是它的调用机制上,一次性把所有工具塞进上下文的机制上出了问题。 1. 上下文预算很快被工具 schema 吃掉。 2. 调用结果直接回填上下文,缺少中间处理层,没有组合性。 于是有人得出结论:CLI 比 MCP 好。 但这其实是一个假象。真正的问题是——能力面(API)和调用面(Agent)之间,需要一个调用层把二者解耦,渐进式披露能力,并且提供可组合性。 最近做 uxc,我反而更确定了这点。 uxc 支持任何带 schema 的接口(MCP / OpenAPI / JSON-RPC),核心不是“换协议”,而是把接口变成一个可渐进披露的 CLI: uxc -h uxc -h uxc key=value uxc '{...}' 能力发现不再是“一次性注入”,而是按需展开。 在这套映射里,MCP 目前仍然是最适合 Agent 的能力暴露协议: 1. schema 暴露是协议强制的,发现路径稳定。 2. description 通常更完整,语义密度更高,模型更容易对齐。 大家经常拿 gh 和 GitHub MCP 比,我自己也比过。 但这个对比其实不太公平。gh 存在时间长,有语料优势,模型天然会用。 你做一个新的服务+ CLI,有这种语料红利吗? 如果没有一个通用调用层,你最后还是会回到老问题: 1. 每个服务维护一套自己的 CLI。 2. 接口一变就要发版迁移。 3. 用户安装和管理成本持续上升。 这和 Agent 时代想要的“能力可组合”是反方向的。 所以我不觉得“CLI 胜出”意味着“MCP 失效”。 更合理的分工是: MCP 负责标准化能力暴露 CLI 负责调用调度 SKILL 负责任务流程 各层负责各层的事情,互相配合。 uxc 0.5.x 这次发布,我主要补了两件基础能力: 1. API key 管理接入 1Password。 2. daemon 模式自动维持 MCP 长连接和 stdio 进程,长期不用自动清理。 然后给了一个 playwright-mcp-cli 的 SKILL 示例(uxc/skills/playwright-mcp-skill)。 这个例子刚好能把三层关系讲清楚: 1. MCP 暴露 Playwright 的能力 schema。 2. uxc 把 endpoint 固化成稳定命令入口,渐进式披露。 3. SKILL 把工具说明和任务流程补齐。 如果要做技术决策,可以简单这么想: 1. 有远程服务想让 Agent 用,需要标准化 API → MCP 是很自然的选项。 2. 有本地程序需要持续交互状态 → MCP stdio 比 daemon+CLI 更干净。 3. 纯一次性命令、不需要状态 → 直接 CLI 就够。 先想清楚三层,再选协议。 去年大家一窝风吹 MCP,我说有点过了。 现在一窝风踩,我又觉得踩过了。 AI 对工具的调用体系,不会靠一个协议就能解决的,需要时间来拼装演化。 1. 从 MCP 到 SKILL:关于 Agent 扩展机制的思考 x.com/jolestar/statu… 2. 从 MCP 到 SKILL(II):把调用层补齐 x.com/jolestar/statu… 3. CLI Is All You Need x.com/mfranz_on/stat…
中文
10
8
91
11.9K
Ray Xiao
Ray Xiao@_RayXiao·
@esmeeezy the beef in the bear doesn't really catch your eyes obviously
English
1
0
1
93
Ray Xiao
Ray Xiao@_RayXiao·
Raw honesty, as always:) I've also thought a lot about how managing a fund differs from managing a regular corporate company. The real complexity hits when a 1–5 person fund tries to scale to 10–20+ people. A talent consulting firm once me told me that some VC management is bit similar as a real estate brokerage model. But great VC firms don't buy into that model—they don't operate as a loose collection of solo agents. To stay competitive long term, your team needs to evolve into full-stack players: absorbing knowledge across a wide range of sectors, sourcing opportunities/scouting, building genuine resonance with founders, being a great salesperson to amplify the brand, and delivering value to portcos even some times just purely as "emotional value" etc etc). The hardest part is figuring out how to build an environment where these "full-stack generals" actually want to stick around and play for the same jersey long term. My takeaway is that to build an enduring fund, the small core group driving the power law must share similar values (which touches on everything you mentioned). "Smartness" here goes way beyond raw IQ it includes a whole spectrum of soft skills. Furthermore, building a positive investment feedback loop in VC is notoriously hard. You won’t know if today’s decision was right in 6 or 12 months. (Even in crypto, where people love to use short term token prices as market validation which is often an illusion). This exact lack of short term validation proves your point: good management is a massive long term edge for a VC firm. When feedback takes 5+ years, your real competitive advantage is the soft power your team builds through time, patience, and shared experience.
English
0
1
3
893
Ray Xiao retweetledi
Lucas
Lucas@OnchainLu·
agentic commerce is accelerating. so here's a breakdown of the current landscape:
Lucas tweet media
English
88
76
514
144.9K
Ray Xiao retweetledi
Alpin Yukseloglu
Alpin Yukseloglu@0xalpo·
new collab from @paradigm and @OpenAI: evmbench is a benchmark and agent harness for exploiting smart contract bugs a few months ago, the best models found <20% of critical, fund-draining @Code4rena bugs in our benchmark. today they find > 70%
OpenAI@OpenAI

Introducing EVMbench—a new benchmark that measures how well AI agents can detect, exploit, and patch high-severity smart contract vulnerabilities. openai.com/index/introduc…

English
69
89
665
179.7K
Ray Xiao retweetledi
Jeremy Allaire - jda.eth / jdallaire.sol
Agent credit onchain Powered by USDC and Circle Gateway
Sumeet (chaos time)@_sumeetc

agent credit on Ethereum is already here and it’s cross-chain @Ch40sChain's Credit Studio enables AI agents with ERC-8004 reputation to autonomously obtain cross-chain USDC liquidity as reputation on one chain unlocks liquidity on another no human approval no pre-bridging no pre-funded wallet on the destination chain

English
31
29
179
20.4K
Ray Xiao retweetledi
X Layer
X Layer@XLayerOfficial·
The Agentic Economy is live on X Layer. ERC-8004 unlocks onchain identity, verifiable reputation, and native discovery for AI agents. Build the next wave of trustless AI on the New Money Chain.
X Layer tweet media
English
41
41
137
29.3K