0xVE

10/29 币安被盗事件又给全行业敲了一次警钟。被攻破的不一定是代码。所有的创始人、开发团队，甚至是运营、HR等等看似离代码很远的岗位，每个人都可能是黑客的突破口。 2024年11月，我们面试过一个前端开发人员。我们收到的是中文简历，中文名字。履历里面记录了他在Bybit, 阿里巴巴等工作经历。在面试的时候，我们发现他的口音是韩式英文，并且不会说中文。我们要求他打开摄像头，摄像头那端，他的像素模糊，画面质感像15年前的摄像头。他带着头戴式耳机，有刘海，和下面截图的这个人非常相似。 我问他会不会说中文，他说不会。我问他是哪里人，他说是日本人。然后我问他在中国和新加坡的工作经历，他含糊其辞。几轮交流下来，我们怀疑他是不是朝鲜黑客，于是结束了面试。从他的简历和背景来看，造假的概率很高。 事情过去了接近两年，今天看到这张照片恍然大悟。 也在此给所有的项目方提醒：面试一定要求开摄像头，PDF不要轻易下载。一定要让全公司的同事都有安全防范意识，不要留下弱点让黑客攻击。 最后，希望加密社区团结一致。我们都在这里支持币安 @cz_binance @heyibinance @zachxbt

Congrats Prof. Huang! Before any self-improvement kicks in, the harness alone — open weights, zero training — is already at 71.97 on OSWorld 2.0 and 83.75 on Agents' Last Exam. Frontier territory, from architecture rather than parameters. RSI then adds on top of that. Two separate results in one table, and the quieter one is doing a lot of work: how you structure the agent is now worth as much as which model you put inside it. Excited to see where the causal line goes next.

~90% chance of a 25bps hike at 2pm. So today's print is already in the price. The real trade is Warsh's guidance, one-and-done, or higher-for-longer as the base case? Credit and BTC don't care about 25bps. They care what funding costs look like for the next year. Watch the path. Not the hike.

Clarity Act failed. 49–50. Needed 60. Not close. BTC dipped. Timeline moved on to the Fed. So stop refreshing the Senate floor. The bill didn’t give you clarity, and waiting for Congress won’t either.

Fed hike tomorrow looks ~90%. 25 bps is basically priced. Bonds already moved. Lending costs are next. Stocks can still pretend it’s “in the price.” So the rate hike isn’t the story anymore. Don’t wait for it.

Most AI progress is still “a higher score on the same exam.” Recursive self-improvement is different. The system doesn’t just answer harder questions, it starts improving the methods that make it smarter. If Shanghai AI Lab’s roadmap is even half right, the scarce thing won’t be another benchmark to brag about. It’ll be judgment: when to let that loop keep running… and when to cut power.

Always building. Always delivering.

景甜不知道，孙哥口中的每一个“好”，都在暗中标好了价格。混娱乐圈的是人中精，景甜以为对孙哥能手拿把掐，却不知道孙哥是币圈的魔。道高一尺，魔高一丈啊，姑娘。

The wait is over. More good news is on the way.

Claude @claudeai is way too expensive. I’m paying $100/month for Max 5x, and I’ve already hit the usage limit twice this month — with no coding, no website building, and no image generation. Just text editing and relatively short documents. If $100/month can’t reliably cover basic professional workflows, it’s hard to justify the cost. For a startup, this is simply not sustainable. @DarioAmodei

RWA was once the most hated, most abandoned asset type. Now it is the most crowded trade in the industry. I have seen a few cycles. The only thing that never changes is that the narrative changes. We tokenized fixed-income debt and T-bills in 2021 and 2022. No users. No investors. Pressure from the SEC, the CFTC and the IRS at the same time. I still remember an investor telling me, "I don't think crypto needs RWA or T-bills. I don't see a use case." I told my co-founders once: when the government starts supporting RWA, we have a shot at being a unicorn. They said that day would never come. They left the industry. We dissolved the company. Five years later, the SEC and CFTC issued joint guidance in March. Nasdaq got approval for tokenized equities in March, NYSE in April. The day came. I am not working on RWA anymore. Here is why. We were first, and we still lost the first-mover advantage. In RWA, first mover is not about market share. It is about who was in the room when the rules got written. Losing it means losing the institutional network, and that is much harder to rebuild than a product. Small builders still send me questions. They are the same questions we had five years ago. The user profile is still mismatched. From tokenized T-bills as the only use case in 2022, to institutions such as market makers and liquid funds trading tokenized stocks without off-ramping in 2026 — that is one big step forward. The next step will also be institutional. Why? Because Institutions have the resources to absorb policy change, and one thing most people underestimate: tax. Right now, the same asset gets two different tax treatments depending on the wrapper. The IRS even has a name for these things, "dual classification assets," which means they have identified the problem without answering it. Digital assets are property, so wash sale rules do not apply. Tokenized securities are securities, so they do apply to stock. That gap is the trade. Most of today's short-term interest in RWA is arbitrage sitting inside it. When the gap closes, the trade either dies or becomes the whole market. What will we do? We have found our role in it. Just wait.

Bitcoin is near $63,000, about half its October 2025 high. Semiconductor ETFs pulled in roughly $21B this year. Spot Bitcoin ETFs saw about $3.1B of net outflows. The capital did not disappear. It went to AI. No one person and no one company reverses a flow like that. @saylor Michael Saylor has carried Bitcoin on his back for six years. He calls Bitcoin superior collateral, and Strategy has been borrowing against it since 2022. The real debate is narrower and more interesting. His model is corporate — preferred securities, deep overcollateralization, institutional balance sheets. What is missing is the same utility for everyone else. Collateral is not leverage for its own sake. It is what turns an asset into a financial system. Treasuries are the most boring asset in the world and also the most rehypothecated. That is not a coincidence. Does it add risk? Yes. 2022 proved badly structured Bitcoin credit can blow up. The answer is not avoiding credit. It is real overcollateralization, transparent LTVs, and no rehypothecation. Bitcoin does not need one more advocate. It needs a credit market. Bitcoin as a pure store of value has a ceiling. Bitcoin as collateral for a working credit market does not. That is an engineering problem, and engineering problems get solved faster with more hands on them. #Bitcoin

My little habit is listening to interviews with business leaders during my workout. Today it was Elon Musk with the editor-in-chief of The Economist. I nearly dropped a dumbbell. She opens so smoothly. "Can we talk about Europe? A place that you are very, very focused on." Then, very politely: judging by your posts, you seem a little concerned. Elon does not take the exit. "I support the normal people. What you call the far right, falsely." That's the part I loved. Two people, zero agreement, nobody blinking, both fully aware the camera is running. The full interview is here youtube.com/watch?v=XuoqKYxD…

Zuckerberg is one of the great founders of our generation. That is what makes this story worth studying. Meta built FAIR in 2013, two years before OpenAI existed. PyTorch came from that lab, and most frontier models are still trained on it. Llama made open weights a real thing. Meta was not late to AI. Meta was earlier than almost anyone. Then in 2021 the company renamed itself after a bold bet. $88B of Reality Labs losses later, it has paid $14.3B for half of Scale AI and cut 600 people from Superintelligence Labs months after record packages brought them in. Those actions look like a student making up work after he realizes he got it wrong. Everyone makes mistakes. The good ones learn fast. Finding the gold and mining it are different skills. The prospector rarely ends up owning the mine. Today's open-weight release is the right instinct. I am watching closely because I learn from the greatest.