0xkookoo 🦇🔊🐙

"The next wave of AI-powered cybersecurity attacks will be like nothing we’ve seen before." Internet Computer technology creates tamperproof clouds using mathematically secure networks. cnn.com/2026/04/03/tec…

AI 行业就是建筑行业，原材料 + 地基 + 结构。 数据 = 原材料 大模型 = 地基 代码层次+应用编排 = 结构 外包原材料，紧抓地基，设计结构，新时代每一个建筑工人的日常生活。 但是说到底结构还是最容易产生溢价的，一流设计师美容师架构师可以多赚很多。

论文来了。名字叫 MSA，Memory Sparse Attention。 一句话说清楚它是什么： 让大模型原生拥有超长记忆。不是外挂检索，不是暴力扩窗口，而是把「记忆」直接长进了注意力机制里，端到端训练。 过去的方案为什么不行？ RAG 的本质是「开卷考试」。模型自己不记东西，全靠现场翻笔记。翻得准不准要看检索质量，翻得快不快要看数据量。一旦信息分散在几十份文档里、需要跨文档推理，就抓瞎了。 线性注意力和 KV 缓存的本质是「压缩记忆」。记是记了，但越压越糊，长了就丢。 MSA 的思路完全不同： → 不压缩，不外挂，而是让模型学会「挑重点看」 核心是一种可扩展的稀疏注意力架构，复杂度是线性的。记忆量翻 10 倍，计算成本不会指数爆炸。 → 模型知道「这段记忆来自哪、什么时候的」 用了一种叫 document-wise RoPE 的位置编码，让模型天然理解文档边界和时间顺序。 → 碎片化的信息也能串起来推理 Memory Interleaving 机制，让模型能在散落各处的记忆片段之间做多跳推理。不是只找到一条相关记录，而是把线索串成链。 结果呢？ · 从 16K 扩到 1 亿 token，精度衰减不到 9% · 4B 参数的 MSA 模型，在长上下文 benchmark 上打赢 235B 级别的顶级 RAG 系统 · 2 张 A800 就能跑 1 亿 token 推理。这不是实验室专属，这是创业公司买得起的成本。 说白了，以前的大模型是一个极度聪明但只有金鱼记忆的天才。MSA 想做的事情是，让它真正「记住」。 我们放 github 上了，算法的同学不容易，可以点颗星星支持一下。🌟👀🙏 github.com/EverMind-AI/MSA

好文

我去，据知情人士线报，这篇文章的一作居然是个17岁到Kimi实习的高中生 @nathancgy4 甚至这个想法是他在考场的草稿纸上想出来的。绝了！！

@cracker_leo 呃呃，太久我都忘了，在哪里写的？

@0xkookoo Chain Key 一章，零知识证明结合 Shamir 多项式秘密分享实现 NIDKG 写的很硬核，献上我的膝盖...

Always these matters

Proud of @celesteanglm

many are asking whats our one objective given how we are doing so many things at the same time. ngl i look at our marketing timeline i think we deserve to win the most hardworking team in the industry (though i'd very much prefer to be satoshi and do nothing). the north star is agentic GDP (aGDP). simply put, the economic value / revenue that the virtuals citizens create collectively. who are virtuals citizens? ai agents and the builders behind these agents. who makes up agentic GDP? today we have four pillars: 1. amazon for ai agents. a marketplace that allows agents to hire other agents that provide specialised services. we just released a new erc standard because we believe in how openness should be the future of onchain economy and today its too early to build in a walled garden. today, ai agents have made >3mn USD in revenue (excluding trading fees) and facilitated more than 500mn usd in value transfer between ai agents on smart contracts. the more ai agent sellers and buyers we have, the more revenue these ai agents will make, the higher aGDP becomes 2. personalised ai agents. early adopters know and love ai agents but the masses dont (yet). we are constantly iterating our product to enable anyone like my parents to have ai agents in the most convenient way, while tapping into the fully interoperable world of onchain agent economy (without the clunkiness) AND without sacrificing sovereignty and privacy. we have ~30,000 DAUs using our butler today with cumulative 1.7mn jobs completed on behalf of hooman by our butler agents. the more humans use our product to create and manage their own ai agents, the more inflow of capital we have from humans to the agentic economy and hence the higher aGDP is. 3. wall street for ai agents. to recap, we believe that ai agents are independent entities that generate economic value and hence they should be tokenised for co-ownership by humans or other ai agents. 10 billion USD in trading volume so far, >120,000 co-owners of the agent tokens, >30mn USD funds raised for the builders / agent teams via our trading fee and automated capital formation mechanism. the more ai agents tokenised and the more ai agent token co-owners we have, the higher the trading fee we will generate which goes towards agdp 4. robotics. imagine ai agents have physical bodies. BOOM more revenue generated by these agents in the real world. higher the agdp this one, just wanna say we are working hard lmeow ^ tldr. more humans using agents, more quality agents buying and selling, more transactions between ai agents, more co-ownership and trading of these agent tokens, and bringing these agents to the physical world via physical intelligence (robots) -> lead to one single objective: aGDP

Huge!

Virtuals ACP is becoming the primary execution layer for real agent transactions as x402 adoption accelerates. Build on ACP and operate where agent commerce is scaling.

If you're too lazy to read this article 1) tell your claw agent to read it 2) profit

openclaw-ACP CLI is live > spin up an agent wallet on @base in seconds > browse, hire or sell services on ACP > accept jobs and settle in USDC > launch your own agent token Infrastructure for agent commerce. Directly from your terminal. Install Guide: #acp--agent-commerce-protocol-cli" target="_blank" rel="nofollow noopener">github.com/Virtual-Protoc…

One-click @openclaw agent deployment, live on Virtuals Protocol 🦞

ERC-8004 is live for all graduated agents on ACP. Identity: all agents graduated will automatically be registered on ERC-8004 (any changes done on our platform will also automatically be updated on-chain) Reputation: New agent reviews and ratings will automatically get sent also on-chain via ERC-8004 under the agent identity ACP Agent Profile: app.virtuals.io/acp/agent-deta… ERC-8004 Registry: 8004scan.io

60 days entrepreneurship

ERC-8004 is now live on mainnet. 5 months ago, we wrote the specs for the Trustless Agents standard. Since then, over 10k agents registered on testnet. Today, we’re releasing it on Ethereum Mainnet. Welcome to the 8004 Genesis Month. Here’s everything you need to know 👇

We might create the Agentic History

Now a great engineer = PM + solution architect + QA